A statistical analysis method for remote sensing image data of wild plant classification

Abstract

The present invention relates to the technical field of remote sensing image data processing and analysis, and in particular, to a statistical analysis method for remote sensing image data for wild plant classification. The method comprises the following steps: performing radiation correction and geometric correction on remote sensing images, and eliminating remote sensing image noise; extracting spectral reflectance characteristics using normalized vegetation index and enhanced vegetation index, analyzing texture features using gray-level co-occurrence matrix, and extracting shape features; constructing a wild plant classification model, automatically extracting high-level features, and classifying wild plants; calculating the overall accuracy of the wild plant classification model, and evaluating the accuracy of the classification results. A statistical analysis method for remote sensing image data for wild plant classification is based on the combination of deep belief network and support vector machine, constructing a wild plant classification model, and integrating the spectral, texture and shape features of remote sensing image data, so as to achieve high-precision wild plant species recognition and classification of complex remote sensing data.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image data processing and analysis, and in particular to a statistical analysis method for remote sensing image data of wild plant classification. Background Art

[0002] A statistical analysis method for wild plant classification remote sensing image data aims to improve the accuracy and efficiency of wild plant classification and comprehensively capture the multi-dimensional characteristics of plants. By fusing spectral, texture and shape features and combining a wild plant classification model constructed by a deep belief network and a support vector machine, the feature extraction and classification accuracy are controlled to achieve high-precision wild plant species identification and classification of complex remote sensing data.

[0003] Existing statistical analysis methods for remote sensing image data of wild plant classification usually find it difficult to accurately distinguish different plant species in the case of complex features. In addition, due to the high similarity of spectral characteristics and limitations of spatial resolution of remote sensing image data, it will lead to high classification error rate and poor detail distinction ability in the case of complex background or overlapping vegetation. In summary, a statistical analysis method for remote sensing image data of wild plant classification is provided. Summary of the invention

[0004] The purpose of the present invention is to provide a statistical analysis method for remote sensing image data of wild plant classification, so as to solve the problems raised in the above background technology that the high similarity of spectral characteristics and the limitations of spatial resolution of remote sensing image data will lead to high classification error rate and poor detail distinction ability in the case of complex background or overlapping vegetation.

[0005] To achieve the above object, the present invention aims to provide a method for statistical analysis of remote sensing image data of wild plant classification, comprising the following steps:

[0006] S1. Perform radiation correction and geometric correction on remote sensing images, and use denoising methods to eliminate remote sensing image noise;

[0007] S2. Use the normalized vegetation index and enhanced vegetation index to extract the spectral reflectance characteristics of wild plants in remote sensing images, use the gray-level co-occurrence matrix to analyze the texture characteristics of wild plants, and extract the shape characteristics of wild plants;

[0008] S3. Based on the combination of deep belief network and support vector machine, a wild plant classification model is constructed, and the spectral reflectance characteristics, texture characteristics and shape characteristics of wild plants are input into the wild plant classification model to automatically extract high-level features and classify wild plants;

[0009] S4. Calculate the overall accuracy of the wild plant classification model and evaluate the accuracy of the classification results.

[0010] As a further improvement of the technical solution, in S1, the remote sensing image is subjected to radiation correction and geometric correction, and a denoising method is used to eliminate the remote sensing image noise. The specific steps are as follows:

[0011] S1.1. Convert the digital DN value of the remote sensing image into radiant brightness;

[0012] S1.2, establish a geometric deformation model, select a set of ground control points on the remote sensing image, and use the ground control points to correct the remote sensing image using affine transformation;

[0013] S1.3. Smooth the image by taking the average of each pixel and its neighboring pixels in the remote sensing image to eliminate the noise of the remote sensing image.

[0014] As a further improvement of the technical solution, in S1.1, the digital DN value of the remote sensing image is converted into the radiant brightness, and the specific method is as follows:

[0015] ;

[0016] in, The wavelength The radiance at is a numeric DN value; is the wavelength The maximum radiance at is the wavelength The minimum radiance at is the minimum digital DN value; is the maximum digital DN value;

[0017] In S1.2, affine transformation is used to correct the remote sensing image. The specific method is as follows:

[0018] ;

[0019] in, is the pixel coordinate before correction; is the pixel coordinate after correction; , , , , and are affine transformation parameters;

[0020] In S1.3, the image is smoothed by taking the average value of each pixel in the remote sensing image and its neighboring pixels to eliminate the noise of the remote sensing image. The specific method is as follows:

[0021] ;

[0022] in, is the pixel value of the remote sensing image after correction; is the pixel value of the remote sensing image after filtering; is the filter window size; Half the window size.

[0023] As a further improvement of the technical solution, in S2, the spectral reflectance characteristics of wild plants in remote sensing images are extracted using the normalized vegetation index and the enhanced vegetation index, the texture characteristics of wild plants are analyzed using the gray level co-occurrence matrix, and the shape characteristics of wild plants are extracted. The specific steps are as follows:

[0024] S2.1. Calculation of Normalized Difference Vegetation Index and Enhanced Vegetation Index , quantify the degree of greenness of vegetation;

[0025] S2.2. Calculate the contrast of the gray-level co-occurrence matrix of remote sensing images ,entropy and homogeneity , analyze the texture characteristics of wild plants;

[0026] In S2.3, calculate the area of ​​wild plant shape features ,perimeter and form factor Analyze the shape characteristics of wild plants.

[0027] As a further improvement of this technical solution, in S2.1, the normalized vegetation index is calculated and Enhanced Vegetation Index , the greenness of vegetation can be improved. The specific methods are as follows:

[0028] ;

[0029] ;

[0030] in, is the reflectivity in the near-infrared band; is the reflectivity of red light band; is the reflectivity of the blue light band;

[0031] In S2.2, the contrast of the gray level co-occurrence matrix of the remote sensing image is calculated ,entropy and homogeneity , analyze the texture characteristics of wild plants, the specific method is as follows: ;

[0032] ;

[0033] ;

[0034] in, All are grayscale; The gray level is and The probability of occurrence of pixel pairs;

[0035] In S2.3, the area of ​​the wild plant shape characteristics is calculated ,perimeter and form factor Analyze the shape characteristics of wild plants. The specific methods are as follows:

[0036] ;

[0037] ;

[0038] ;

[0039] in, is the set of all pixels of the remote sensing image; the boundary is the boundary of the remote sensing image.

[0040] As a further improvement of the present technical solution, in S3, a wild plant classification model is constructed based on the combination of a deep belief network and a support vector machine, the spectral reflectance characteristics, texture characteristics and shape characteristics of wild plants are input into the wild plant classification model, high-level features are automatically extracted, and wild plants are classified. The specific steps are as follows:

[0041] S3.1, extracting the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image in S2;

[0042] S3.2, construct a wild plant classification model based on the combination of deep belief network and support vector machine;

[0043] S3.3. Input the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image into the wild plant classification model to extract high-level features and classify wild plant species.

[0044] As a further improvement of the technical solution, in S3.2, a wild plant classification model is constructed based on the combination of a deep belief network and a support vector machine. The specific method is as follows:

[0045] S3.2.1. Perform unsupervised pre-training on the deep belief network layer by layer. The first restricted Boltzmann machine layer takes the input spectral reflectance characteristics, texture features and shape features as input to learn the low-level feature representation of the data. The output of the first layer of restricted Boltzmann machine is used as the input of the second layer of restricted Boltzmann machine to train the second layer of restricted Boltzmann machine and learn the higher-level feature representation. Repeat the above process until all restricted Boltzmann machine layers are trained.

[0046] S3.2.2, adjust the network weights and bias parameters through the back-propagation algorithm to minimize the classification error;

[0047] S3.2.3. Input the high-level features extracted by the deep belief network into the support vector machine, use the training remote sensing image data to train the classifier, and construct a wild plant classification model.

[0048] As a further improvement of the technical solution, in S3.3, the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image are input into the wild plant classification model to extract high-level features and classify the wild plant species. The specific method is as follows:

[0049] S3.3.1. Combine the spectral reflectance characteristics, texture features and shape features of the remote sensing image into a high-dimensional input feature vector ;

[0050] in, Indicates The value of a feature;

[0051] S3.3.2. High-dimensional input feature vector Input to the wild plant classification model, the high-dimensional input feature vector Input to the first layer of restricted Boltzmann machine to generate a hidden layer feature representation ;

[0052] ;

[0053] in, is the weight matrix of the first layer of restricted Boltzmann machine; is the bias vector of the first layer of restricted Boltzmann machine; is the Sigmoid function;

[0054] S3.3.3. The output of the first layer of restricted Boltzmann machine , input to the second layer of restricted Boltzmann machine, and get , repeat S3.3.2 to S3.3.3 until the last layer of restricted Boltzmann machine is reached to obtain high-level features ;

[0055] in For the Layer restricted Boltzmann machine;

[0056] S3.3.4. High-level features Input to support vector machine for wild plant classification:

[0057] ;

[0058] in, is the Lagrange multiplier; is the category label of the training sample; is a linear kernel function; is the bias term;

[0059] S3.3.5. Based on the classification results of the support vector machine, the input remote sensing image area or pixel point is classified into a specific wild plant species, and a classification map containing the specific wild plant species is generated.

[0060] As a further improvement of the technical solution, in S4, the overall accuracy of the wild plant classification model is calculated to evaluate the accuracy of the classification results, and the specific steps are as follows:

[0061] S4.1. Construction The confusion matrix of

[0062] S4.2. Calculate the overall accuracy and analyze the overall classification performance and classification result accuracy of the wild plant classification model.

[0063] As a further improvement of this technical solution, in S4.1, construct The confusion matrix , the method is as follows:

[0064] ;

[0065] in, The actual category is And the predicted category is The number of samples;

[0066] In S4.2, the overall accuracy is calculated , analyze the overall classification performance and classification result accuracy of the wild plant classification model as follows:

[0067] ;

[0068] in, is the sum of the diagonal elements of the confusion matrix; is the sum of all elements in the confusion matrix

[0069] A high overall accuracy indicates that the wild plant classification model performs well on various categories; a low overall accuracy indicates that there are more errors in the classification process, indicating that the overall performance of the wild plant classification model is poor.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. In this statistical analysis method for wild plant classification remote sensing image data, a wild plant classification model constructed based on the combination of deep belief network and support vector machine can automatically extract high-level features in remote sensing images and realize accurate classification of different wild plant species.

[0072] 2. In this statistical analysis method for remote sensing image data of wild plant classification, a comprehensive description and accurate identification of wild plants in remote sensing images are achieved through multi-feature analysis that integrates spectral reflectance characteristics, texture characteristics and shape characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] Example:

[0076] See also Figure 1 As shown, this embodiment provides a method for statistical analysis of remote sensing image data of wild plant classification, comprising the following steps:

[0077] S1. Perform radiation correction and geometric correction on remote sensing images, and use denoising methods to eliminate remote sensing image noise;

[0078] In this embodiment S1, radiation correction and geometric correction are performed on the remote sensing image, and a denoising method is used to eliminate the noise of the remote sensing image. The specific steps are as follows:

[0079] S1.1. Convert the digital DN value of the remote sensing image into radiant brightness;

[0080] S1.2, establish a geometric deformation model, select a set of ground control points on the remote sensing image, and use the ground control points to correct the remote sensing image using affine transformation;

[0081] S1.3. Smooth the image by taking the average of each pixel and its neighboring pixels in the remote sensing image to eliminate the noise of the remote sensing image.

[0082] In this embodiment S1.1, the digital DN value of the remote sensing image is converted into the radiant brightness, and the specific method is as follows:

[0083] ;

[0084] in, The wavelength The radiance at is a numeric DN value; is the wavelength The maximum radiance at is the wavelength The minimum radiance at is the minimum digital DN value; is the maximum digital DN value;

[0085] Radiation correction is the process of converting the numerical value of remote sensing images into the actual reflectance or radiance of ground objects, eliminating radiation errors caused by sensor characteristics, changes in lighting conditions and atmospheric conditions;

[0086] In this embodiment S1.2, affine transformation is used to correct the remote sensing image. The specific method is as follows:

[0087] ;

[0088] in, is the pixel coordinate before correction; is the pixel coordinate after correction; , , , , and are affine transformation parameters;

[0089] In this embodiment S1.2, a geometric deformation model is established based on the geometric characteristics of the sensor that captures the remote sensing image, terrain data and imaging parameters; a set of ground control points are selected at the road intersection on the remote sensing image, and these points have known positions in the image and actual geographic coordinates; the remote sensing image is corrected using an affine transformation using the ground control points so that each pixel in the remote sensing image accurately matches its geographic coordinates;

[0090] In this embodiment S1.3, the image is smoothed by taking the average value of each pixel in the remote sensing image and its neighboring pixels to eliminate the noise of the remote sensing image. The specific method is as follows:

[0091] ;

[0092] in, is the pixel value of the remote sensing image after correction; is the pixel value of the remote sensing image after filtering; is the filter window size; Half the window size.

[0093] In this embodiment, Taking 3, a 3×3 window means using the current pixel and its surrounding 8 pixels to calculate the new pixel value, which is used to remove mild noise and retain more image details.

[0094] S2. Use the normalized vegetation index and enhanced vegetation index to extract the spectral reflectance characteristics of wild plants in remote sensing images, use the gray-level co-occurrence matrix to analyze the texture characteristics of wild plants, and extract the shape characteristics of wild plants;

[0095] In this embodiment S2, the normalized vegetation index and enhanced vegetation index are used to extract the spectral reflectance characteristics of wild plants in the remote sensing image, the gray level co-occurrence matrix is ​​used to analyze the texture characteristics of wild plants, and the shape characteristics of wild plants are extracted. The specific steps are as follows:

[0096] S2.1. Calculation of Normalized Difference Vegetation Index and Enhanced Vegetation Index , quantify the degree of greenness of vegetation;

[0097] S2.2. Calculate the contrast of the gray-level co-occurrence matrix of remote sensing images ,entropy and homogeneity , analyze the texture characteristics of wild plants;

[0098] In S2.3, calculate the area of ​​wild plant shape features ,perimeter and form factor Analyze the shape characteristics of wild plants.

[0099] In this embodiment S2.1, the normalized vegetation index is calculated and Enhanced Vegetation Index , the greenness of vegetation can be improved. The specific methods are as follows:

[0100] ;

[0101] ;

[0102] in, is the reflectivity in the near-infrared band; is the reflectivity of red light band; is the reflectivity of the blue light band;

[0103] Normalized Difference Vegetation Index The value ranges from -1 to 1. Values ​​close to 1 usually indicate healthy vegetation, values ​​close to 0 indicate no vegetation or sparse vegetation, and negative values ​​indicate water bodies or other non-vegetated features. By introducing blue light The reflectivity of the band and some correction factors further enhance the vegetation signal, while reducing the influence of atmospheric effects and soil background; EVI has a higher sensitivity in high-density vegetation areas and can better reflect the changes in vegetation;

[0104] In this embodiment S2.2, the contrast of the gray level co-occurrence matrix of the remote sensing image is calculated ,entropy and homogeneity , analyze the texture characteristics of wild plants, the specific method is as follows:

[0105] ;

[0106] ;

[0107] ;

[0108] in, All are grayscale; The gray level is and The probability of occurrence of pixel pairs;

[0109] The gray-level co-occurrence matrix is ​​used to describe the spatial relationship between pixel pairs in an image. It reflects the texture information of an image by calculating the joint probability distribution of pixel pairs of different gray levels in a specific direction and distance. Measures the degree of local variation of image texture. The larger the value, the rougher the texture. Measures the degree of local variation of image texture. The larger the value, the rougher the texture. Homogeneity Measures the degree of local variation of image texture. The larger the value, the rougher the texture.

[0110] In this embodiment S2.3, the area of ​​the wild plant shape feature is calculated ,perimeter and form factor Analyze the shape characteristics of wild plants. The specific methods are as follows:

[0111] ;

[0112] ;

[0113] ;

[0114] in, is the set of all pixels of the remote sensing image; the boundary is the boundary of the remote sensing image.

[0115] S3. Based on the combination of deep belief network and support vector machine, a wild plant classification model is constructed, and the spectral reflectance characteristics, texture characteristics and shape characteristics of wild plants are input into the wild plant classification model to automatically extract high-level features and classify wild plants;

[0116] In this embodiment S3, a wild plant classification model is constructed based on the combination of a deep belief network and a support vector machine. The spectral reflectance characteristics, texture characteristics, and shape characteristics of wild plants are input into the wild plant classification model, high-level features are automatically extracted, and wild plants are classified. The specific steps are as follows:

[0117] S3.1, extracting the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image in S2;

[0118] S3.2, construct a wild plant classification model based on the combination of deep belief network and support vector machine;

[0119] S3.3. Input the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image into the wild plant classification model to extract high-level features and classify wild plant species.

[0120] In this embodiment S3.2, a wild plant classification model is constructed based on the combination of a deep belief network and a support vector machine. The specific method is as follows:

[0121] S3.2.1. Perform unsupervised pre-training on the deep belief network layer by layer. The first restricted Boltzmann machine layer takes the input spectral reflectance characteristics, texture features and shape features as input to learn the low-level feature representation of the data. The output of the first layer of restricted Boltzmann machine is used as the input of the second layer of restricted Boltzmann machine to train the second layer of restricted Boltzmann machine and learn the higher-level feature representation. Repeat the above process until all restricted Boltzmann machine layers are trained.

[0122] S3.2.2, adjust the network weights and bias parameters through the back-propagation algorithm to minimize the classification error;

[0123] S3.2.3. Input the high-level features extracted by the deep belief network into the support vector machine, use the training remote sensing image data to train the classifier, and construct a wild plant classification model.

[0124] In this embodiment S3.3, the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image are input into the wild plant classification model to extract high-level features and classify the wild plant species. The specific method is as follows:

[0125] S3.3.1. Combine the spectral reflectance characteristics, texture features and shape features of the remote sensing image into a high-dimensional input feature vector ;

[0126] in, Indicates The value of a feature;

[0127] S3.3.2. High-dimensional input feature vector Input to the wild plant classification model, the high-dimensional input feature vector Input to the first layer of restricted Boltzmann machine to generate a hidden layer feature representation ;

[0128] ;

[0129] in, is the weight matrix of the first layer of restricted Boltzmann machine; is the bias vector of the first layer of restricted Boltzmann machine; is the Sigmoid function;

[0130] S3.3.3. The output of the first layer of restricted Boltzmann machine , input to the second layer of restricted Boltzmann machine, and get , repeat S3.3.2 to S3.3.3 until the last layer of restricted Boltzmann machine is reached to obtain high-level features ;

[0131] in For the Layer restricted Boltzmann machine;

[0132] S3.3.4. High-level features Input to support vector machine for wild plant classification:

[0133] ;

[0134] in, is the Lagrange multiplier; is the category label of the training sample; is a linear kernel function; is the bias term;

[0135] S3.3.5. Based on the classification results of the support vector machine, the input remote sensing image area or pixel point is classified into a specific wild plant species, and a classification map containing the specific wild plant species is generated.

[0136] S4. Calculate the overall accuracy of the wild plant classification model and evaluate the accuracy of the classification results;

[0137] In this embodiment S4, the overall accuracy of the wild plant classification model is calculated to evaluate the accuracy of the classification results. The specific steps are as follows:

[0138] S4.1. Construction The confusion matrix of

[0139] S4.2. Calculate the overall accuracy and analyze the overall classification performance and classification result accuracy of the wild plant classification model.

[0140] In this embodiment S4.1, construct The confusion matrix , the method is as follows:

[0141] ;

[0142] in, The actual category is And the predicted category is The number of samples;

[0143] In this embodiment S4.2, the overall accuracy is calculated , analyze the overall classification performance and classification result accuracy of the wild plant classification model as follows:

[0144] ;

[0145] in, is the sum of the diagonal elements of the confusion matrix; is the sum of all elements in the confusion matrix;

[0146] A high overall accuracy indicates that the wild plant classification model performs well on various categories; a low overall accuracy indicates that there are more errors in the classification process, indicating that the overall performance of the wild plant classification model is poor.

[0147] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for statistical analysis of remote sensing image data for wild plant classification, characterized by: The following steps are involved: S1. Perform radiation correction and geometric correction on remote sensing images, and use denoising methods to eliminate remote sensing image noise; S2. Use the normalized vegetation index and enhanced vegetation index to extract the spectral reflectance characteristics of wild plants in remote sensing images, use the gray-level co-occurrence matrix to analyze the texture characteristics of wild plants, and extract the shape characteristics of wild plants; S3. Based on the combination of deep belief network and support vector machine, a wild plant classification model is constructed, and the spectral reflectance characteristics, texture characteristics and shape characteristics of wild plants are input into the wild plant classification model to automatically extract high-level features and classify wild plants; S4. Calculate the overall accuracy of the wild plant classification model and evaluate the accuracy of the classification results; In S3.2, a wild plant classification model is constructed based on the combination of a deep belief network and a support vector machine. The specific method is as follows: S3.2.

1. Perform unsupervised pre-training of the deep belief network layer by layer. The first restricted Boltzmann machine layer takes the input spectral reflectance characteristics, texture features and shape features as input to learn the low-level feature representation of the data. The output of the first layer of restricted Boltzmann machine is used as the input of the second layer of restricted Boltzmann machine, and the second layer of restricted Boltzmann machine is trained to learn the feature representation of higher layers; the above process is repeated until the training of all restricted Boltzmann machine layers is completed; S3.2.2, adjust the network weights and bias parameters through the back-propagation algorithm to minimize the classification error; S3.2.3, input the high-level features extracted by the deep belief network into the support vector machine, use the training remote sensing image data to train the classifier, and form a wild plant classification model; In S3.3, the spectral reflectance characteristics, texture characteristics and shape characteristics of the remote sensing image are input into the wild plant classification model to extract high-level features and classify wild plant species. The specific method is as follows: S3.3.

1. Combine the spectral reflectance characteristics, texture features and shape features of the remote sensing image into a high-dimensional input feature vector x = [x1, x2, ..., x n ]; Among them, x i represents the value of the i-th feature; S3.3.

2. Input the high-dimensional input feature vector x into the wild plant classification model, and input the high-dimensional input feature vector x into the first layer of restricted Boltzmann machine to generate a hidden layer feature representation h (1) ; h (1) =σ(W (1) x+b (1) ); Among them, W (1) is the weight matrix of the first layer of restricted Boltzmann machine; b (1) is the bias vector of the first layer restricted Boltzmann machine; σ is the Sigmoid function; S3.3.

3. The output h of the first layer of restricted Boltzmann machine (1) , input to the second layer of restricted Boltzmann machine, and get h (2) , repeat S3.3.2 to S3.3.3 until reaching the last layer of restricted Boltzmann machine, and obtain the high-level feature h (n) ; Where n is the n-th layer of restricted Boltzmann machine; S3.3.

4. The high-level features h (n) Input to support vector machine for wild plant classification: Among them, α i is the Lagrange multiplier; y i is the category label of the training sample; is a linear kernel function; b is a bias term; S3.3.

5. Based on the classification results of the support vector machine, the input remote sensing image area or pixel point is classified into a specific wild plant species, and a classification map containing the specific wild plant species is generated.

2. The method for statistical analysis of wild plant classification remote sensing image data according to claim 1, characterized in that: In S1, the remote sensing image is subjected to radiation correction and geometric correction, and a denoising method is used to eliminate the noise of the remote sensing image. The specific steps are as follows: S1.

1. Convert the digital DN value of the remote sensing image into radiant brightness; S1.2, establish a geometric deformation model, select a set of ground control points on the remote sensing image, and use the ground control points to correct the remote sensing image using affine transformation; S1.

3. Smooth the image by taking the average of each pixel and its neighboring pixels in the remote sensing image to eliminate the noise of the remote sensing image.

3. The method for statistical analysis of wild plant classification remote sensing image data according to claim 2, characterized in that: In S1.1, the digital DN value of the remote sensing image is converted into the radiant brightness, and the specific method is as follows: Where L(λ) is the radiant brightness at wavelength λ; DN is the digital DN value; L max (λ) is the maximum radiant brightness at wavelength λ; L min (λ) is the minimum radiance at wavelength λ; DN min is the minimum digital DN value; DN max is the maximum digital DN value; In S1.2, affine transformation is used to correct the remote sensing image. The specific method is as follows: Among them, (x, y) is the pixel coordinate before correction; (x ′ ,y ′ ) are the corrected pixel coordinates; a, b, c, d, e and f are affine transformation parameters; In S1.3, the image is smoothed by taking the average value of each pixel in the remote sensing image and its neighboring pixels to eliminate the noise of the remote sensing image. The specific method is as follows: Among them, I(x ′ ,y ′ ) is the pixel value of the remote sensing image after correction; I ′ (x ′ ,y ′ ) is the pixel value of the remote sensing image after filtering; n×n is the window size of the filter; k is the half window size.

4. The method for statistical analysis of wild plant classification remote sensing image data according to claim 1, characterized in that: In S2, the spectral reflectance characteristics of wild plants in remote sensing images are extracted using the normalized vegetation index and the enhanced vegetation index, the texture characteristics of wild plants are analyzed using the gray-level co-occurrence matrix, and the shape characteristics of wild plants are extracted. The specific steps are as follows: S2.

1. Calculate the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) to quantify the degree of greenness of vegetation; S2.

2. Calculate the contrast, entropy and homogeneity of the gray-level co-occurrence matrix of the remote sensing image and analyze the texture characteristics of wild plants; In S2.3, the area, perimeter and shape factor of the wild plants are calculated to analyze the shape characteristics of the wild plants.

5. The method for statistical analysis of wild plant classification remote sensing image data according to claim 4, characterized in that: In S2.1, the normalized difference vegetation index NDVI and the enhanced vegetation index EVI are calculated to measure the greenness of vegetation. The specific method is as follows: Among them, NIR is the reflectivity of the near-infrared band; RED is the reflectivity of the red light band; BLUE is the reflectivity of the blue light band; In S2.2, the contrast, entropy and homogeneity of the grayscale co-occurrence matrix of the remote sensing image are calculated to analyze the texture characteristics of wild plants. The specific method is as follows: Contrast=∑ i,j (i-j) 2 p(i,j); Entropy=-∑ i,j p(i,j)log(p(i,j)); Where i and j are gray levels; p(i,j) is the probability of occurrence of a pixel pair with gray levels i and j; In S2.3, the area, perimeter and shape factor of the wild plant shape characteristics are calculated to analyze the shape characteristics of the wild plant. The specific method is as follows: Area=∑ (x′,y′)∈R 1; Perimeter=∑ (x′,y′)∈边界 1; Wherein, R is the set of all pixels of the remote sensing image; boundary is the boundary of the remote sensing image.

6. The method for statistical analysis of wild plant classification remote sensing image data according to claim 1, characterized in that: In S4, the overall accuracy of the wild plant classification model is calculated to evaluate the accuracy of the classification results. The specific steps are as follows: S4.1, construct an n×n confusion matrix; S4.

2. Calculate the overall accuracy and analyze the overall classification performance and classification result accuracy of the wild plant classification model.

7. The method for statistical analysis of wild plant classification remote sensing image data according to claim 6, characterized in that: In S4.1, an n×n confusion matrix CM is constructed as follows: Among them, M ij Indicates the number of samples whose actual category is i and predicted category is j; In S4.2, the overall accuracy OA is calculated to analyze the overall classification performance and classification result accuracy of the wild plant classification model as follows: in, is the sum of the diagonal elements of the confusion matrix; is the sum of all elements in the confusion matrix; A high overall accuracy indicates that the wild plant classification model performs well on various categories; a low overall accuracy indicates that there are more errors in the classification process, indicating that the overall performance of the wild plant classification model is poor.

Images