Method for extracting spatial distribution information of mountain orchid rice producing area based on optical and radar images

Abstract

The invention provides a method for extracting spatial distribution information of a mountain orchid rice producing area based on optical and radar images. The method comprises the following steps: S11, acquiring optical and radar image data of mountain orchid rice; s12, converting an RGB image in the image data of the mountain, orchid and rice into an HSV color space for denoising; s13, performing inversion on the surface temperature of the mountain orchid rice planting concentration area to improve the mountain orchid rice identification precision; s14, using a separation threshold method to screen out an optimal feature set with a relatively high feature category separation degree from the feature categories of the mountain orchid rice planting concentration area; s15, analyzing the optimal feature set by using a mixed pixel decomposition method to obtain a training feature set; and S16, classifying the training feature set by using a random forest model to obtain mountain, orchid and rice space distribution information. According to the method, on the basis of optical and radar images, image denoising and land surface temperature inversion are carried out, so that the mountain, orchid and rice recognition precision is improved, and then the mixed pixel decomposition method and the random forest model are utilized to efficiently obtain mountain, orchid and rice space distribution information.

Description

Technical Field

[0001] The present invention relates to the technical field of Shanlan rice identification, and in particular to a method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images. Background Art

[0002] Shanlan rice is a kind of mountain dry rice unique to the Li nationality in Hainan. It is a precious treasure preserved by the Li nationality in their long-term production practice. Affected by natural conditions, agricultural development and urbanization, the spatial distribution of Shanlan rice has changed dramatically. Therefore, the study of the spatiotemporal distribution and changes of Shanlan rice is of great significance. With the continuous advancement and widespread application of remote sensing technology, its application in spatial dynamic monitoring has been significantly improved. Compared with traditional crop monitoring methods, remote sensing technology has unique advantages such as short time consumption, high efficiency, wide monitoring range, and high temporal continuity and repeatability. However, for most tropical and subtropical areas, it is very difficult to obtain a sufficient number of critical period images of Shanlan rice.

[0003] The imaging data of optical and radar images are not affected by meteorological interference and can respond sensitively to the development of mountain rice plants and changes in soil moisture. They are an important data source for monitoring the growth of mountain rice in foggy areas. Although the existing classification scheme can achieve a high classification accuracy for mountain rice, due to the problem of too many noise points in the image data of optical and radar images, there is still a lack of extraction accuracy. Summary of the invention

[0004] In order to overcome the problem of low precision in extracting the area of ​​mountain rice in complex planting areas such as low mountain and hilly areas, the present invention provides a method for extracting spatial distribution information of mountain rice producing areas based on optical and radar images, comprising the following steps:

[0005] S11, perform cloud mask processing on the GEE (Google Earth Engine) platform, set the images with cloud cover less than the preset cloud cover threshold for synthesis, and obtain the optical and radar image data of Shanlandao;

[0006] S12, converting the RGB image in the image data obtained in step S11 into the HSV color space, using the 8-dimensional space vector information of a single pixel to form a polygonal pixel aggregate, and denoising the image data of the mountain blue rice based on the polygonal pixel aggregate to retain the edge detail characteristics of the mountain blue rice;

[0007] S13, inversion of surface temperature, using a universal single-channel algorithm to invert the surface temperature of the concentrated rice planting area after denoising in step S12, so as to improve the recognition accuracy of rice;

[0008] S14, extracting the characteristic categories of the concentrated mountain rice planting area in step S13, wherein the characteristic categories include spectral bands, vegetation indexes, water body indexes, and terrain and texture characteristics, and then using a separation threshold method to screen out a preferred characteristic set of concentrated mountain rice planting areas whose characteristic category separation degree is greater than a preset separation degree threshold;

[0009] S15, using a fully constrained least squares mixed pixel decomposition method, analyzing the optimal feature set in step S14, obtaining an abundance map of land types that are easily confused with the spectrum of mountain blue rice, and integrating it into the classification system as a training feature set;

[0010] S16, constructing a random forest model, classifying the rice training feature set in step S15, and obtaining the spatial distribution information of Shanlan rice.

[0011] Furthermore, in step S12, forming a polygonal pixel aggregate using the 8-dimensional space vector information of a single pixel includes the following steps:

[0012] S21, expressing a single pixel 8-dimensional space vector as (u, v, h, s, l, r, g, b), which is used to express the pixel point of the image data of mountain rice, wherein (u, v) represents the spatial position vector, (h, s, l) is the HSV color feature, and (r, g, b) is the RGB color feature. Based on the (h, s, l, r, g, b) color feature, the number of local center seed points n is calculated;

[0013] S22, the number of cluster centers calculated by S21 is the number n of local centers of the pixels of the mountain blue rice image data, the number of edges of the polygonal pixel aggregate is taken, and the standard spacing of the aggregate is calculated;

[0014] S23, taking the cluster center point as the local center, searching for the color feature distance and spatial distance between all pixels in the vicinity within a range of twice the standard distance and the cluster center, and updating the pixel label information;

[0015] S24, start iterative calculation for each pixel point, after repeated iterations, when the distance error between each cluster center and the centers around it converges to 0.1, the iteration is terminated, and the polygonal pixel cluster is determined.

[0016] Furthermore, in step S12, the denoising of the image data of the mountain rice based on the polygonal pixel aggregate comprises the following steps:

[0017] S31. Set the search boxΦ i Similar to the fixed-size box, the search box contains multiple polygonal pixel aggregates;

[0018] S32, traverse the polygon pixel set, compare its central pixel j with the pixel to be processed i, and compare the similarity of all pixels of i and j to obtain the weight w of pixel j ij ;

[0019] S33, finally, weighted average is performed on all polygon pixel aggregates in the search box to obtain the pixel estimation value of the denoised image at point i

[0020] S34, outputting the denoised image data of the mountain rice.

[0021] Furthermore, in step S14, the separation threshold method is used to screen and obtain the optimal feature set:

[0022] S41, use SEaTH's JM (Jeffries-Matusita) distance to judge the separability of feature categories. Its value range is [0, 2]. Its value is close to 0, which means that the two categories have almost no difference in a certain feature; its value is 2, which means that the two categories can be completely distinguished in a certain feature.

[0023] S42, combining field survey data with image data of concentrated Shanlan rice planting areas to assist in visual interpretation;

[0024] S43, calculating the separation degree of the feature categories, retaining the two feature categories with the largest separation degree, including repeated feature categories, and obtaining the optimal feature set for the concentrated Shanlan rice planting area.

[0025] Further, in step S15, the step of analyzing the Shanlan rice planting concentrated area data set is:

[0026] S51, taking the optimal feature set of the concentrated area of ​​mountain rice planting as the input data set, performing principal component analysis on the feature variables through minimum noise separation to achieve data dimensionality reduction and estimate image noise points;

[0027] S52, calculating a pure pixel index according to the image noise;

[0028] S53, using n-dimensional visualization tools to select pure pixels of various types;

[0029] S54, obtaining vegetation abundance maps of mountain rice, other crops and vegetables based on fully constrained least squares mixed pixel decomposition;

[0030] S55. Use vegetation abundance maps as an aid to distinguish land types that are prone to spectral confusion.

[0031] Furthermore, in step S16, the random forest model construction steps are:

[0032] S61. Use the Gini index to perform feature selection when splitting nodes. The Gini index represents the probability of a random sample in a sample set being misclassified. The expression of the Gini index Gini(D) of the set D is constructed as follows:

[0033]

[0034] In the formula, K is the number of categories in the training samples; is the probability that a randomly selected sample from set D belongs to category K;

[0035] S62. If set D is divided into two parts, D1 and D2, according to whether feature A takes a certain value α, then under the condition of feature A, the Gini index Gini(D,A) of set D is expressed as:

[0036]

[0037] Where |D| represents the number of samples in set D, |D1| represents the number of samples in set D1, and |D2| represents the number of samples in set D2;

[0038] S63. The average impurity reduction method is used to evaluate the importance of the training feature set, and its expression is:

[0039]

[0040] Among them, N tree is the number of decision trees in the random forest model; Gini t (D) Gini t (D, A) are the Gini index of set D before and after the t-th decision tree is partitioned by feature A.

[0041] Furthermore, in step S43, the step of evaluating the importance of the training feature set by using the average impurity reduction method is:

[0042] S71. Use the average impurity reduction method to calculate the importance of each feature in the training feature set in the random forest model, and sort them in descending order;

[0043] S72, when ranking according to the importance of each feature, the first feature is selected for the first time, the first two features are selected for the second time, and so on, a random forest model of 20 different feature combinations of a single phase and 100 different feature combinations of time series images can be obtained;

[0044] S73. Calculate the out-of-bag (OOB) data respectively, and determine the optimal feature combination after comprehensively considering the accuracy and complexity of the random forest model, so as to reduce the model complexity while ensuring the classification accuracy.

[0045] Further, in step S16, the method for extracting the spatial distribution information of mountain blue rice is:

[0046] S81, using the random forest model to randomly select N samples from the training feature set with replacement as training samples through the bootstrap method;

[0047] S82, N trte N times of sample extraction and training can be obtained tree A decision tree model;

[0048] S83. Randomly select M at each node of each decision tree. try Features and internal node division;

[0049] S84. Integrate all decision tree model results and use majority voting to determine the final classification result of the spatial distribution information of mountain rice.

[0050] The beneficial effects of the present invention are as follows:

[0051] The present invention relates to a method for extracting spatial distribution information of mountain rice production areas based on optical and radar images. First, image data of mountain rice optical and radar images are obtained on a GEE platform, so that the planting area and growth information of mountain rice can be obtained timely and accurately; RGB images in the mountain rice image data are converted to HSV color space for denoising, which well retains the edge detail characteristics of mountain rice, is conducive to improving the inversion accuracy, and provides a basis for subsequent extraction and classification of mountain rice, thereby improving the accuracy of extracting spatial distribution information of mountain rice; secondly, a universal single-channel algorithm is used to invert the surface temperature of the mountain rice planting concentrated area after denoising, thereby further improving the recognition accuracy of mountain rice; finally, a least squares mixed pixel decomposition method and a random forest model are used for classification to obtain the spatial distribution information of mountain rice. The present invention first performs denoising on the image before extracting and classifying mountain rice, thereby improving the image recognition accuracy, thereby improving the accuracy of extracting spatial distribution information of mountain rice. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0053] Figure 1 The present invention is a flow chart of a method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images. DETAILED DESCRIPTION

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0055] The present invention provides a method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images, comprising the following steps:

[0056] S11. Perform cloud mask processing on the GEE (Google Earth Engine) platform, set the images with cloud cover less than the preset cloud cover threshold for synthesis, and combine the main phenological period of Shanlan single-season rice in the study area to obtain the optical and radar image data of Shanlan rice. Optical remote sensing data is easily affected by weather factors such as precipitation and fog, and it is difficult to meet the spatial distribution extraction of Shanlan rice in rainy areas; the space-borne synthetic aperture radar SAR (Synthetic Aperture Radar) can achieve target observation by actively emitting electromagnetic waves and recording the characteristics of surface scattered echoes. It can continuously image the ground without being affected by weather conditions and has all-day and all-weather observation capabilities; therefore, according to the phenological period of Shanlan rice, the optical and radar data are synthesized in time series, and the synthesized image data is obtained by data fusion based on the synthesized data, which can solve the problem that the southern region is affected by geographical location, fog, etc., the optical data is seriously missing and the single image has low accuracy in crop extraction.

[0057] S12, converting the RGB image in the image data obtained in step S11 to the HSV color space, using the 8-dimensional space vector information of a single pixel to form a polygonal pixel aggregate, and denoising the image data of the mountain blue rice based on the polygonal pixel aggregate to retain the edge detail characteristics of the mountain blue rice. This algorithm uses 8-dimensional vectors as pixel-level information expression, which can better retain the edge detail characteristics of the image in terms of noise resistance. The algorithm can realize the image denoising function under different devices and different interferences.

[0058] In this embodiment, the paddy field filled with water and mixed with mountain blue rice plants in the RGB image is enhanced and displayed as blue-purple. The water body is displayed as blue, the forest is displayed as dark green, and the buildings, bare land and other objects are displayed as gray or khaki. Using RGB images, mountain blue rice fields can be highlighted on the map, thereby distinguishing them from other types of objects. In the RGB color space, changes in lighting conditions may cause changes in the values ​​of different objects on the same color channel, which brings difficulties to the selection of thresholds. The HSV color space divides color information into three components: hue (H), saturation (S) and brightness (V). In image processing, hue corresponds to the color information of the object and can be used to distinguish different types of objects. Saturation reflects the purity and brightness of the color and can be used to distinguish the materials and features of different objects. Brightness represents the brightness information of the image and can provide information about the reflectivity and radiance of the object. In the HSV color system, the V channel is relatively insensitive to changes in lighting, making the threshold selection more stable. Separating brightness information from color information can better adapt to changes in lighting conditions and improve the robustness of recognition. Mapping the RGB channels in the image separately and converting them to the HSV color system for recognition can make full use of the information in the image to improve the accuracy of recognition.

[0059] In step S12, forming a polygonal pixel aggregate using the 8-dimensional space vector information of a single pixel includes the following steps:

[0060] S21. Represent a single pixel 8-dimensional space vector as (u, v, h, s, l, r, g, b) to express the pixel point of the image data of mountain rice, where (u, v) represents the spatial position vector, (h, s, l) is the HSV color feature, and (r, g, b) is the RGB color feature. Based on the (h, s, l, r, g, b) color feature, the steps of calculating the number of local center seed points n are as follows:

[0061] Create k points as initial centroids, where k is a random number in the number of image pixels (width*height).

[0062] When the cluster assignment result of any point changes:

[0063] For each data point in the feature set, the Euclidean distance from the centroid to the color features of all data points is calculated, and the data point is assigned to the cluster with the closest distance.

[0064] For each cluster, calculate the mean of all points in the cluster and use the mean as the new centroid.

[0065] When the number of clusters is less than k:

[0066] For each cluster, calculate the total error and continue clustering on the given cluster;

[0067] Calculate the total error of splitting the cluster into two, and select the cluster with the smallest error for the split operation.

[0068] S22, the number of cluster centers calculated by S21 is the number of local centers of the pixel points of the mountain rice image data n, the number of pixels of the image is M=R×C, the number of edges of the polygonal pixel aggregate is taken, and the standard spacing of the aggregate is calculated. The number of edges of the polygonal pixel aggregate ranges from 4 to 8, which can be set according to the final effect. When the value is less than 6, it will cause the pixels to be not smooth enough. When the value is greater than 6, the amount of calculation becomes large. Based on this, this embodiment selects 6 for calculation, and the obtained effect is relatively good. Then the standard spacing calculation formula of the aggregate is:

[0069]

[0070] Where M represents the number of image pixels, R represents the number of image rows, C represents the number of image columns, s represents the standard distance of the aggregate, M represents the total number of pixels, and n represents the number of edges of the aggregate.

[0071] S23, taking the cluster center point as the local center, searching for the color feature distance and spatial distance between all pixels in the vicinity within a range of twice the standard distance and the cluster center, and updating the pixel label information;

[0072] S24, start to repeat the calculation for each pixel point, after repeated iterations, when the distance error between each cluster center and the center around it converges to 0.1, the iteration is terminated, and the polygon pixel set is determined. The polygon pixel set used by the algorithm, as the basic processing method of the image, can not only improve the effect of the denoising algorithm, but also improve the operational efficiency of the enhancement algorithm. When the convergence value is greater than 0.1, the clustering effect will be poor, and when the convergence value is less than 1, it will stop converging. Therefore, in this embodiment, the iteration is terminated when the distance error converges to 0.1. After the iteration is terminated, the final category center, category label, category range and other information can be obtained.

[0073] Furthermore, in step S12, the denoising of the image data of the mountain rice based on the polygonal pixel aggregate comprises the following steps:

[0074] S31. Set the search boxΦ i Similar to the fixed-size box, the search box contains multiple polygonal pixel aggregates;

[0075] S32, traverse the polygon pixel set, compare its central pixel j with the pixel to be processed i, and compare the similarity of all pixels of i and j to obtain the weight w of pixel j ij ;

[0076] S33, finally, weighted average is performed on all polygon pixel aggregates in the search box to obtain the pixel estimation value of the denoised image at point i

[0077] S34, outputting the denoised image data of the mountain rice.

[0078] In this embodiment, the purpose of setting the search box is to get rid of the defect of slow global search speed. In order to improve the speed of image processing, the image to be processed is divided into several small blocks, and then searched in the search box. After the search is completed, filtering is performed to change its current value, and the original value is changed to a new value, that is, So we need to further estimate w ij Update and get the value of each search box.

[0079] S13, inversion of surface temperature, using a universal single-channel algorithm to invert the surface temperature of the concentrated rice planting area after denoising in step S12, so as to improve the recognition accuracy of rice;

[0080] In this embodiment, the physiological and biochemical effects of different mountain blue rice growth processes can produce a significant warming or cooling effect on the LST (ground surface temperature) of the rice field. LST data, especially LST data during the critical growth period of mountain blue rice, can effectively improve the recognition accuracy of mountain blue rice. Existing studies have shown that the single-channel algorithm is more advantageous than the split-window algorithm for LST data inversion. Therefore, the present invention uses a universal single-channel algorithm to invert LST data.

[0081] S14, extracting the feature categories of the concentrated mountain rice planting area in step S13, the feature categories including spectral bands, vegetation index, water body index, and terrain and texture features, and then using the separation threshold method to screen out the preferred feature set of the concentrated mountain rice planting area with a feature category separation degree greater than a preset separation degree threshold; selecting the feature categories to be used in the process of identifying the mountain rice, considering that the mountain rice has plant characteristics and requires water to grow, if all the selected feature categories are used for classification, not only will the classification time be prolonged, but the classification accuracy will also be affected due to data redundancy. In order to reduce data redundancy, the separation threshold method is used for screening.

[0082] Furthermore, in step S14, the separation threshold method is used to screen and obtain the optimal feature set:

[0083] S41, use SEaTH's JM (Jeffries-Matusita) distance to judge the separability of feature categories. Its value range is [0, 2]. Its value is close to 0, which means that the two categories have almost no difference in a certain feature; its value is 2, which means that the two categories can be completely distinguished in a certain feature.

[0084] S42. Combine field survey data with image data of concentrated rice planting areas to assist in visual interpretation of rice, forest land, water area, construction land and other land types;

[0085] S43, calculating the separation degree of the feature categories, retaining the two feature categories with the largest separation degree, including repeated feature categories, and obtaining the optimal feature set for the concentrated Shanlan rice planting area.

[0086] The calculation formula of separation degree is:

[0087] J = 2(1-e -B )

[0088]

[0089] Among them, J is the separation; B is the Bhattacharyya distance, which is used to calculate the distance between factors; m1 and m2 are the means of a certain characteristic distribution of two different samples; σ1 and σ2 are the standard deviations of a certain characteristic distribution of two different samples.

[0090] S15, using a fully constrained least squares mixed pixel decomposition method, analyzing the optimal feature set in step S14, obtaining an abundance map of land types that are easily confused with the spectrum of mountain blue rice, and integrating it into the classification system as a training feature set;

[0091] Further, in step S15, the step of analyzing the Shanlan rice planting concentrated area data set is:

[0092] S51, taking the optimal feature set of the concentrated area of ​​mountain rice planting as the input data set, performing principal component analysis on the feature variables through minimum noise separation to achieve data dimensionality reduction and estimate image noise points;

[0093] S52, calculating the pure pixel index according to the image noise, wherein the hyper parameters in the calculation process, including the number of iterations, the iteration unit and the threshold coefficient, are set to 10000, 250 and 2.5 respectively;

[0094] S53, using n-dimensional visualization tools to select pure pixels of various types;

[0095] S54, obtaining vegetation abundance maps of mountain rice, other crops and vegetables based on fully constrained least squares mixed pixel decomposition;

[0096] S55. Use vegetation abundance maps as an aid to distinguish land types that are prone to spectral confusion.

[0097] In this embodiment, a fully constrained least squares mixed pixel decomposition method is used to obtain abundance maps of land types (vegetables and other crops) that are easily spectrally confused with mountain rice, and the maps are integrated into the classification system as training feature sets. The above method is used to obtain abundance maps of mountain rice, other crops, and vegetable bases in different periods. The vegetation abundance obtained by mixed pixel decomposition can better distinguish the three land types that are easily spectrally confused.

[0098] S16, constructing a random forest model, classifying the rice training feature set in step S15, and obtaining the spatial distribution information of Shanlan rice. The random forest algorithm is an ensemble learning method that uses decision trees as basic classifiers and combines bagging ensemble learning theory with random subspace methods.

[0099] Furthermore, in step S16, the random forest model construction steps are:

[0100] S61. When constructing a decision tree in a random forest, the Gini index is used to select features when splitting nodes. The Gini index represents the probability that a random sample in a sample set is misclassified. The smaller the Gini index, the higher the purity of the set and the smaller the probability of being misclassified; conversely, the less pure the set. The expression of the Gini index Gini(D) of the constructed set D is:

[0101]

[0102] In the formula, K is the number of categories in the training samples; is the probability that a randomly selected sample from set D belongs to category K;

[0103] S62. If set D is divided into two parts, D1 and D2, according to whether feature A takes a certain value α, then under the condition of feature A, the Gini index Gini(D,A) of set D is expressed as:

[0104]

[0105] Where |D| represents the number of samples in set D, |D1| represents the number of samples in set D1, and |D2| represents the number of samples in set D2;

[0106] S63. It can be seen from this that when constructing a random forest model, the greater the reduction in the Gini index after a feature partition, the greater the degree to which the set becomes purer after the partition, and the more important the feature is in the model. Therefore, the average impurity reduction method is used to evaluate the importance of the training feature set, and its expression is:

[0107]

[0108] Among them, N tree is the number of decision trees in the random forest model; Gini t(D), Gini t (D, A) are the Gini indices of the set D before and after being partitioned by the feature A in the t-th decision tree respectively.

[0109] Furthermore, in step S43, the steps of evaluating the importance of the training feature set by using the method of average impurity reduction are as follows:

[0110] S71. Calculate the importance of each feature in the training feature set in the random forest model by using the method of average impurity reduction, and sort them in descending order;

[0111] S72. When sorting according to the importance of each feature, select the first feature for the first time, select the first two features for the second time, and so on, to obtain a random forest model with 20 different feature combinations for single-temporal images and 100 different feature combinations for time-series images;

[0112] S73. Calculate their out-of-bag (OOB) data respectively, and determine the optimal feature combination after comprehensively considering the accuracy and complexity of the random forest model, so as to reduce the model complexity under the condition of ensuring the classification accuracy.

[0113] Furthermore, in step S16, the method for extracting the spatial distribution information of upland rice is as follows:

[0114] S81. Use the random forest model to randomly select N samples from the training feature set with replacement by the bootstrap method. Therefore, about 37% of the samples are not drawn, and these samples are called out-of-bag (OOB) data, which can be used to evaluate the performance of the random forest model;

[0115] S82. After N tree times of sample extraction and training, N tree decision tree models can be obtained;

[0116] S83. Randomly select M try features (M try < M, where M is the total number of features) at each node of each decision tree and perform internal node partitioning;

[0117] S84. Integrate the results of all decision tree models, and use the majority voting method to determine the final classification result of the upland rice spatial distribution information.

[0118] In this embodiment, the OOB error is the most stable when the number of decision trees is 100. Therefore, in this paper, Ntree is finally set to 100, and Mtry is set to the square root of the total number of input features.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The present invention provides a method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images, characterized in that: The following steps are involved: S11, perform cloud mask processing on the GEE (Google Earth Engine) platform, set the images with cloud cover less than the preset cloud cover threshold for synthesis, and obtain the optical and radar image data of Shanlandao; S12, converting the RGB image in the image data obtained in step S11 into the HSV color space, using the 8-dimensional space vector information of a single pixel to form a polygonal pixel aggregate, and denoising the image data of the mountain blue rice based on the polygonal pixel aggregate to retain the edge detail characteristics of the mountain blue rice; S13, inversion of surface temperature, using a universal single-channel algorithm to invert the surface temperature of the concentrated rice planting area after denoising in step S12, so as to improve the recognition accuracy of rice; S14, extracting the characteristic categories of the concentrated mountain rice planting area in step S13, wherein the characteristic categories include spectral bands, vegetation indexes, water body indexes, and terrain and texture characteristics, and then using a separation threshold method to screen out a preferred characteristic set of concentrated mountain rice planting areas whose characteristic category separation degree is greater than a preset separation degree threshold; S15, using a fully constrained least squares mixed pixel decomposition method, analyzing the optimal feature set in step S14, obtaining an abundance map of land types that are easily confused with the spectrum of mountain blue rice, and integrating it into the classification system as a training feature set; S16, constructing a random forest model, classifying the rice training feature set in step S15, and obtaining the spatial distribution information of Shanlan rice.

2. The method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images according to claim 1, characterized in that, in step S12, forming a polygonal pixel aggregate using 8-dimensional spatial vector information of a single pixel comprises the following steps: S21, expressing a single pixel 8-dimensional space vector as (u, v, h, s, l, r, g, b), which is used to express the pixel point of the image data of mountain rice, wherein (u, v) represents the spatial position vector, (h, s, l) is the HSV color feature, and (r, g, b) is the RGB color feature. Based on the (h, s, l, r, g, b) color feature, the number of local center seed points n is calculated; S22, the number of cluster centers calculated by S21 is the number n of local centers of the pixels of the mountain blue rice image data, the number of edges of the polygonal pixel aggregate is taken, and the standard spacing of the aggregate is calculated; S23, taking the cluster center point as the local center, searching for the color feature distance and spatial distance between all pixels in the vicinity within a range of twice the standard distance and the cluster center, and updating the pixel label information; S24, start iterative calculation for each pixel point, after repeated iterations, when the distance error between each cluster center and the centers around it converges to 0.1, the iteration is terminated, and the polygonal pixel cluster is determined.

3. The method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images according to claim 1 is characterized in that, in step S12, denoising the Shanlan rice image data based on polygonal pixel aggregates comprises the following steps: S31. Set the search boxΦ i Similar to the fixed-size box, the search box contains multiple polygonal pixel aggregates; S32, traverse the polygon pixel set, compare its central pixel j with the pixel to be processed i, and compare the similarity of all pixels of i and j to obtain the weight w of pixel j ij ; S33, finally, weighted average is performed on all polygon pixel aggregates in the search box to obtain the pixel estimation value of the denoised image at point i S34, outputting the denoised image data of the mountain rice.

4. The method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images according to claim 1, characterized in that, in step S14, the method for obtaining the optimal feature set by screening using the separation threshold method is: S41, use SEaTH's JM (Jeffries-Matusita) distance to judge the separability of feature categories. Its value range is [0, 2]. Its value is close to 0, which means that the two categories have almost no difference in a certain feature; its value is 2, which means that the two categories can be completely distinguished in a certain feature. S42, combining field survey data with image data of concentrated Shanlan rice planting areas to assist in visual interpretation; S43, calculating the separation degree of the feature categories, retaining the two feature categories with the largest separation degree, including repeated feature categories, and obtaining the optimal feature set for the concentrated Shanlan rice planting area.

5. The method for extracting spatial distribution information of Shanlan rice producing areas based on optical and radar images according to claim 1 is characterized in that, in step S15, the step of analyzing the Shanlan rice concentrated planting area data set is: S51, taking the optimal feature set of the concentrated area of ​​mountain rice planting as the input data set, performing principal component analysis on the feature variables through minimum noise separation to achieve data dimensionality reduction and estimate image noise points; S52, calculating a pure pixel index according to the image noise; S53, using n-dimensional visualization tools to select pure pixels of various types; S54, obtaining vegetation abundance maps of mountain rice, other crops and vegetables based on fully constrained least squares mixed pixel decomposition; S55. Use vegetation abundance maps as an aid to distinguish land types that are prone to spectral confusion.

6. The method for extracting spatial distribution information of mountain rice producing areas based on optical and radar images according to claim 1, characterized in that, in step S16, the random forest model construction step is: S61. Use the Gini index to perform feature selection when splitting nodes. The Gini index represents the probability of a random sample in a sample set being misclassified. The expression of the Gini index Gini(D) of the set D is constructed as follows: In the formula, K is the number of categories in the training samples; is the probability that a randomly selected sample from set D belongs to category K; S62. If set D is divided into two parts, D1 and D2, according to whether feature A takes a certain value α, then under the condition of feature A, the Gini index Gini(D,A) of set D is expressed as: in, |D| represents the number of samples in set D, |D1| represents the number of samples in set D1, and |D2| represents the number of samples in set D2; S63. The average impurity reduction method is used to evaluate the importance of the training feature set, and its expression is: Among them, N tree is the number of decision trees in the random forest model; Gini t (D) Gini t (D, A) are the Gini index of the set D before and after the t-th decision tree is partitioned by feature A.

7. The method for extracting spatial distribution information of mountain rice producing areas based on optical and radar images according to claim 6, characterized in that, in step S43, the step of evaluating the importance of the training feature set by using the method of reducing the average impurity is: S71. The importance of each feature in the training feature set in the random forest model is calculated using the average impurity reduction method, and the features are sorted in descending order. S72, when ranking according to the importance of each feature, the first feature is selected for the first time, the first two features are selected for the second time, and so on, a random forest model of 20 different feature combinations of a single phase and 100 different feature combinations of time series images can be obtained; S73. Calculate the out-of-bag (OOB) data respectively, and determine the optimal feature combination after comprehensively considering the accuracy and complexity of the random forest model, so as to reduce the model complexity while ensuring the classification accuracy.

8. The method for extracting spatial distribution information of mountain rice producing areas based on optical and radar images according to claim 7, characterized in that, in step S16, the method for extracting spatial distribution information of mountain rice is: S81, using the random forest model to randomly select N samples from the training feature set with replacement as training samples through the bootstrap method; S82, N tree N times of sample extraction and training can be obtained tree A decision tree model; S83. Randomly select M at each node of each decision tree. try Features and internal node division; S84. Integrate all decision tree model results and use majority voting to determine the final classification result of the spatial distribution information of mountain rice.

Images