Accurate crop extraction method based on remote sensing of unmanned aerial vehicle

Abstract

The invention discloses an accurate crop extraction method based on unmanned aerial vehicle remote sensing, and belongs to the field of agricultural resource remote sensing measurement technology application. A standardized flight experiment is carried out on specific crops through equipment such as a multi-rotor unmanned aerial vehicle, and a standard remote sensing image at a fixed height is obtained; standard preprocessing such as geometric distortion correction, graying and unified cutting is carried out on the image; determining a crop periodic scale in the preprocessed image, and extracting a gray level co-occurrence matrix and a wavelet decomposition texture feature parameter; calculating each texture feature weight, and establishing crop comprehensive texture feature index parameters; and an identification rule is established according to the index parameters and the texture feature quantity, and accurate extraction of crop types and spatial distribution areas is realized. The method overcomes the problems that traditional satellite remote sensing is low in spectral resolution, different in spectrum and the like, compared with a traditional method, the method has the advantages of being high in operability, standard in process, low in cost, high in spatial resolution, high in recognition precision and the like, crop information can be efficiently and accurately extracted, and powerful support is provided for related agricultural research and production.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of agricultural resource remote sensing measurement, and particularly relates to a crop precise extraction method based on unmanned aerial vehicle remote sensing. BACKGROUND

[0002] Correctly identifying and extracting the types of surface crops and scientifically grasping the spatial distribution of surface crops are a basic work of agricultural resource measurement. The crop image data obtained at different altitudes by using remote sensing means is a main data source for monitoring and counting ground crop information, and can realize qualitative and quantitative research on the geometric shape, spatial position and development change of surface crops. Among them, the unmanned aerial vehicle remote sensing technology has a series of characteristics such as strong real-time, high spatial resolution and convenient load replacement, which can effectively make up for the problems of long acquisition period, low spatial resolution and limited load of satellite remote sensing. In addition, in the specific application of small and medium scale, the unmanned aerial vehicle remote sensing has greater advantages, can quickly, mobile and accurately obtain the ground information, and combined with the ground measured data, realizes the monitoring task of the ground information in a certain area. Through the combination of unmanned aerial vehicle remote sensing image acquisition and ground manual investigation, a large amount of information such as the type, area, coverage and spatial distribution of crops can be obtained in time and space, which can provide data support for guiding agricultural production, forestry resource planning and management, pest monitoring, and subsequent ecological environment evaluation and soil and water conservation research, and has great scientific research value and practical application significance.

[0003] For the actual measurement and identification of crops, the traditional method must be obtained by investigation means, and the area is not determined. By using the traditional remote sensing classification method, the phenomenon of same object different spectrum and different object same spectrum cannot be avoided, which greatly affects the identification accuracy of crops in remote sensing image. SUMMARY

[0004] Based on the above deficiencies, the application provides a crop precise extraction method based on unmanned aerial vehicle remote sensing, which can overcome the problems of low spectral resolution and large data volume of traditional satellite remote sensing image, and avoid the problems of same object different spectrum and different object same spectrum when using spectral reflectance for classification.

[0005] The technology adopted by the application is as follows: a crop precise extraction method based on unmanned aerial vehicle remote sensing, comprising the following steps:

[0006] Step 1: Acquisition of standard crop drone remote sensing images: Conduct standardized flight experiments on specific crop species using a multi-rotor drone, CCD high-definition camera, handheld GPS, compass, mobile computer, theodolite, and metal calibration plate. Determine the drone's flight altitude, flight parameters, and flight path, obtain calibrated drone remote sensing images at a fixed altitude, and perform standardized measurements on homogeneous remote sensing images of specific crops under these calibration conditions.

[0007] Step 2: Standardized preprocessing of drone images: Use the coordinate data of the calibration points to perform geometric distortion correction on the drone-calibrated remote sensing images at a fixed altitude in Step 1. Perform geometric correction on the homogeneous remote sensing images of specific crops at that altitude. Convert the geometric correction results to grayscale, and crop the grayscale images of the specific crop remote sensing images to a uniform size to form a preprocessed result of a uniform standard size.

[0008] Step 3: Standardized extraction of crop texture feature parameters: Generate random seed points for the preprocessed image of the specific crop in Step 2. Calculate statistical features within different sizes and determine the periodic scale of the crop with the seed points as the center. Generate 20 random mask regions based on the sizes, and extract gray-level co-occurrence matrix texture features from the specific crop image within the region, as well as wavelet texture features at different decomposition levels.

[0009] Step 4. Establishment of comprehensive texture feature index parameters for crops: Calculate the texture feature weights of the gray-level co-occurrence matrix texture feature quantities and wavelet decomposition texture feature quantities in the 20 random mask areas in step 3, standardize each feature parameter based on the 20 samples, and then calculate the mean of a certain parameter after standardization. Then, based on the mean of different types, calculate the proportion of the mean to all values ​​as the weight, and on this basis establish the comprehensive texture feature index parameters for crops;

[0010] Step 5. Identification and identification of crop species: Using the standardized UAV flight plan in step 1, perform standardized UAV coordinate positioning aerial photography flights on the entire area to be measured. Establish a window to traverse the standard pre-processed images of UAV remote sensing images and calculate different types of texture feature quantities. Establish identification rules based on the range of texture feature quantities and the comprehensive texture feature index parameters of crops to achieve accurate extraction of specific crops from the remote sensing images of the entire area and final determination of the spatial distribution area.

[0011] Further, the step 1 determines the flight height, flight parameters and flight route of the unmanned aerial vehicle, specifically: selecting a determined square region on the map, measuring the area and average altitude of the region, and determining the latitude and longitude information of the boundary points of the region; dividing the region into multiple square overlapping sub-regions, and recording the center points of the sub-regions; calculating the optimal flight height according to the side length of the sub-region and the field of view angle of the CCD lens of the unmanned aerial vehicle, and calculating the latitude and longitude coordinates of the center points of the sub-regions in combination with the latitude and longitude coordinates of the corner points of the region and the geometric position relationship of the sub-regions; placing a metal calibration plate above the center points of the sub-regions, setting the actual flight height of the unmanned aerial vehicle according to the measured altitude of the region, setting the flight route and hovering shooting position of the unmanned aerial vehicle, and making the unmanned aerial vehicle fly along the straight path of the center points of the sub-regions and sequentially measure and shoot.

[0012] Further, the step 2 of uniformly sizing and cropping the specific crop remote sensing image grayscale map is to crop the grayscale remote sensing image by measuring the position of the boundary points of the region, and delete the ground objects and images in the non-crop region.

[0013] Further, the step 3 of determining the periodic scale of the crop is specifically: setting the number of random seed points and the variable neighborhood radius in the preprocessed result image for simulation, calculating the mean, variance and standard deviation of the neighborhood for each seed point, averaging the first-order statistical quantities of all neighborhoods under each cycle to obtain a function of the statistical characteristic quantity changing with the neighborhood radius, and the double of the minimum radius value corresponding to the stable state of the function image is the periodic structure scale value corresponding to the crop.

[0014] Further, the step 3 of extracting the gray level co-occurrence matrix texture feature quantity of the specific crop image in the region includes extracting contrast, correlation, consistency and angular second moment.

[0015] Further, the step 3 of using coiflet-1 wavelet basis function to perform 4-level wavelet decomposition processing on the sub-image in the random training sample region, and calculating the norm and energy value wavelet decomposition feature quantity of the low frequency, horizontal high frequency, vertical high frequency and diagonal high frequency according to the wavelet decomposition coefficients at different levels.

[0016] Further, the step 4 of calculating the texture feature weight is specifically: performing summation normalization processing on each texture feature parameter extraction result data set, calculating the mean of all sample points after the summation normalization processing of each texture feature quantity, calculating the sum of the normalized means of different types of texture features, and taking the ratio of each texture feature normalized mean to the sum as the weight of the texture feature in all texture features.

[0017] Further, the step 5 extracts different types of texture features of the random sample of the unknown crop to be identified, calculates the comprehensive texture feature index parameter of the crop to be identified according to the weight of different types of crops in the known database, and calculates the difference ratio between the crop to be identified and the known crop, and classifies the crop to be identified as the type of the known crop with the closest result of 0.

[0018] The method of the present application can be trained for specific crop types, enabling accurate identification and extraction of specific crops on a large scale. It has strong operability, standardized processing flow, clear parameter extraction process, and clear crop comprehensive texture feature index parameter. Compared with traditional field research methods, the method of the present application saves a lot of manpower, material resources and time cost in identifying the type and spatial distribution of specific crops. Compared with traditional satellite remote sensing image feature extraction and classification methods, the unmanned aerial vehicle remote sensing method of the present application has strong maneuverability, low cost and high spatial resolution; the present application greatly improves the identification accuracy of specific crops and effectively improves the accuracy of crop area, boundary and spatial distribution feature extraction. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The flight path map of the unmanned aerial vehicle of the present application. DETAILED DESCRIPTION

[0020] In order to more clearly illustrate the purpose, technical scheme and advantages of the present application, the specific embodiments of the present application will be further described in detail below in combination with the drawings and specific implementations. The following examples are used to illustrate the present application, but not to limit the scope of the present application.

[0021] A crop precise extraction method based on unmanned aerial vehicle remote sensing, the specific steps are as follows:

[0022] Example 1

[0023] Step 1: This step involves using multi-rotor unmanned aerial vehicle, handheld GPS, compass, theodolite, notebook computer, data connection line, metal calibration plate and other devices and apparatus. It mainly involves selecting a reasonable unmanned aerial vehicle remote sensing flight height, and carrying out unmanned aerial vehicle remote sensing monitoring of specific crop types at this height, obtaining standardized unmanned aerial vehicle remote sensing images of crops at this height, and the specific steps are as follows:

[0024] First, for a specific crop type, such as corn crops planted in a certain area, select a certain square area on Google Map, and measure the area of the square area as Area (in this example, the side length of the square area is 100 meters, and the corresponding Area = 10000m 2), the average height h of the region is measured by the theodolite, and the latitude and longitude information of the four boundary points of the region on Google Maps is determined, wherein the latitude and longitude of the upper left corner point point-1 are latitude-1 and longitude-1 respectively, the latitude and longitude of the upper right corner point point-2 are latitude-2 and longitude-2 respectively, the latitude and longitude of the lower left corner point point-3 are latitude-3 and longitude-3 respectively, and the latitude and longitude of the lower right corner point point-4 are latitude-4 and longitude-4 respectively.

[0025] Secondly, according to the actual area Area of the region, the region is divided into 9 square overlapping sub-regions Zone-1, Zone-2, Zone-3, Zone-4, Zone-5, Zone-6, Zone-7, Zone-8, Zone-9 covering the entire region, and the center points P1, P2, P3, P4, P5, P6, P7, P8, P9 of each sub-region are recorded, ensuring that the square sub-regions adjacent in the horizontal and vertical directions have half the area of overlap. For this example, the length of each square sub-region is 50m, and each sub-region is numbered according to the following distribution. Secondly, according to the length L=50m of each square sub-region, the field of view A of the UAV CCD lens is set to 90 degrees, and the corresponding optimal flight height H=0.5L / tan(0.5A)=25m of the UAV remote sensing is calculated.

[0026] Thirdly, according to the latitude and longitude coordinates of the four corner points point-1, point-2, point-3, point-4 of the measurement region, combined with the geometric position relationship of different square sub-regions, the latitude and longitude coordinates of the center point of each square sub-region are calculated.

[0027] Fourthly, a metal calibration plate is placed above the center point of each square sub-region, and the actual flight height H' of the UAV is set to H'+h according to the height h of the measurement region. Fourthly, the flight route and hovering photographing position of the UAV are set so that the UAV flies along the straight line path P1-P2-P3-P4-P5-P6-P7-P8-P9 and sequentially measures each square region by UAV remote sensing. The UAV high-definition remote sensing image of each square sub-region is taken vertically for 3 seconds, and the UAV high-definition remote sensing images of all square sub-regions are saved as I1, I2, I3, I4, I5, I6, I7, I8, I9, thereby realizing the acquisition of standardized UAV high-definition remote sensing images of square sub-regions of the entire specific crop measurement region.

[0028] Step 2: This step involves the standardization of the entire specific crop measurement area of the square sub-region standardization of the UAV high-definition remote sensing image, the specific steps are as follows:

[0029] First, using the four corner points of the entire measurement area and the latitude and longitude coordinates of each square sub-region center position point, respectively, the geometric distortion correction of each square sub-region UAV high-definition remote sensing image is carried out, and Point-1, P1, P2, P4, P5 are used to correct the geometric distortion of I1, and the geometric distortion correction result is saved as I1'; P1, P2, P3, P4, P5, P6 are used to correct the geometric distortion of I2, and the geometric distortion correction result is saved as I2'; Point-2, P2, P3, P5, P6 are used to correct the geometric distortion of I3, and the geometric distortion correction result is saved as I3'; P2, P3, P5, P6, P8, P9 are used to correct the geometric distortion of I4, and the geometric distortion correction result is saved as I4'; P1, P2, P3, P4, P5, P6, P7, P8, P9 are used to correct the geometric distortion of I5, and the geometric distortion correction result is saved as I5'; P1, P2, P4, P5, P7, P8 are used to correct the geometric distortion of I6, and the geometric distortion correction result is saved as I6'; Point-3, P4, P5, P7, P8 are used to correct the geometric distortion of I7, and the geometric distortion correction result is saved as I7'; P4, P5, P6, P7, P8, P9 are used to correct the geometric distortion of I8, and the geometric distortion correction result is saved as I8'; Point-4, P5, P6, P8, P9 are used to correct the geometric distortion of I9, and the geometric distortion correction result is saved as I9'.

[0030] Second, using the corrected square sub-region coordinate point position, the image splicing based on the geometric position point is carried out on the adjacent square sub-region, and all the nine square sub-regions are spliced into a complete square remote sensing image covering the entire crop measurement area, and the image splicing result is saved as I.

[0031] Third, load I in MATLAB software, and record the red image component R, green image component G, and blue image component B of I image in the form of matrix, use the formula (R+G+B) / 3 to carry out the gray processing of the entire area remote sensing image based on the color component, and save the gray processing result as I'. Fourth, using Point-1, Point-2, Point-3, Point-4 position points to crop the entire crop area gray remote sensing image I', and crop and delete other non-integral square crop area and image, and save the cropping result as I".

[0032] Step 3: This step involves the standardization extraction of different types of texture features for the specific crop pretreatment result image in step 2, the specific steps are as follows:

[0033] Firstly, load I" in MATLAB software, get the row number Row and column number Col of I", which represent the number of pixels in the rows and columns of the image I", respectively. Since the UAV measurement area is a square, Row = Col.

[0034] Secondly, set 50 random seed points in the I" image, and set a variable radius r neighborhood radius for simulation, where r gradually increases from 1 pixel to 0.5*Row pixels.

[0035] Thirdly, loop through each seed point, a total of 50 times, each loop generates random points according to the increasing neighborhood radius, each random seed point corresponds to a neighborhood radius, then calculates the mean, variance and standard deviation of the neighborhood, which are three first-order statistical features. Average the first-order statistical results of each neighborhood under all loops, that is, get the function of the statistical feature with the neighborhood radius. The function image is stable after reaching a certain value, and the double of the minimum radius value corresponding to the stable state is the periodic structure scale value Scale of the crop in the remote sensing image.

[0036] Fourthly, select 20 random points in I", and determine a square region with a side length of Scale as the training sample region for each random point as the center. Based on the second-order combined conditional probability density, the gray level co-occurrence matrix texture feature extraction is performed on the sub-image in each training sample region. Specifically, the contrast is extracted and the contrast extraction results of all 20 sample points are saved as the contrast dataset X1, the correlation is extracted and the correlation extraction results of all 20 sample points are saved as the correlation dataset X2, the consistency is extracted and the consistency extraction results of all 20 sample points are saved as the consistency dataset X3, and the angular second moment is extracted and the angular second moment extraction results of all 20 sample points are saved as the angular second moment dataset X4.

[0037] In the fifth step, the sub-images in each random training sample region generated in the fourth step are processed by 4-level wavelet decomposition using coiflet-1 wavelet basis function. The low-frequency norm wavelet decomposition feature quantity and the low-frequency energy value wavelet decomposition feature quantity are calculated according to the low-frequency decomposition coefficients after the fourth-level wavelet decomposition, and the extraction results of all 20 sample points are saved as the low-frequency norm wavelet decomposition feature quantity dataset X5 and the low-frequency energy value wavelet decomposition feature quantity dataset X6 respectively. The horizontal high-frequency norm wavelet decomposition feature quantity and the horizontal high-frequency energy value wavelet decomposition feature quantity are calculated according to the horizontal high-frequency decomposition coefficients after the fourth-level wavelet decomposition, and the extraction results of all 20 sample points are saved as the horizontal high-frequency norm wavelet decomposition feature quantity dataset X7 and the horizontal high-frequency energy value wavelet decomposition feature quantity dataset X8 respectively. The vertical high-frequency norm wavelet decomposition feature quantity and the vertical high-frequency energy value wavelet decomposition feature quantity are calculated according to the vertical high-frequency decomposition coefficients after the fourth-level wavelet decomposition, and the extraction results of all 20 sample points are saved as the vertical high-frequency norm wavelet decomposition feature quantity dataset X9 and the vertical high-frequency energy value wavelet decomposition feature quantity dataset X10 respectively. The diagonal high-frequency norm wavelet decomposition feature quantity and the diagonal high-frequency energy value wavelet decomposition feature quantity are calculated according to the diagonal high-frequency decomposition coefficients after the fourth-level wavelet decomposition, and the extraction results of all 20 sample points are saved as the diagonal high-frequency norm wavelet decomposition feature quantity dataset X11 and the diagonal high-frequency energy value wavelet decomposition feature quantity dataset X12 respectively.

[0038] In step 4, the different types of texture feature standardized extraction results in step 3 are used to establish a comprehensive crop texture feature index parameter, and the specific steps are as follows:

[0039] First step, for the 20 random sample areas in step 3, the sum of the normalized data set of each texture feature parameter is calculated, the mean of all sample points after the sum of the normalized data of each texture feature is calculated, the mean of the sum of the normalized data of the contrast texture feature is saved as x1, the mean of the sum of the normalized data of the correlation texture feature is saved as x2, the mean of the sum of the normalized data of the consistency texture feature is saved as x3, the mean of the sum of the normalized data of the angular second moment texture feature is saved as x4, the mean of the sum of the normalized data of the low frequency norm wavelet decomposition feature is saved as x5, the mean of the sum of the normalized data of the low frequency energy value wavelet decomposition feature is saved as x6, the mean of the sum of the normalized data of the horizontal high frequency norm wavelet decomposition feature is saved as x7, the mean of the sum of the normalized data of the horizontal high frequency energy value wavelet decomposition feature is saved as x8, the mean of the sum of the normalized data of the vertical high frequency norm wavelet decomposition feature is saved as x9, the mean of the sum of the normalized data of the vertical high frequency energy value wavelet decomposition feature is saved as x10, the mean of the sum of the normalized data of the diagonal high frequency norm wavelet decomposition feature is saved as x11, and the mean of the sum of the normalized data of the diagonal high frequency energy value wavelet decomposition feature is saved as x12.

[0040] Second step, the sum of the normalized mean of different types of texture features of the 20 samples in the previous step is calculated, and the ratio of the calculation result to the sum of the normalized mean of each texture feature in the previous step is calculated again, and the ratio of the calculation result is taken as the weight of each texture feature in all texture features, the weight of the contrast texture feature is saved as q1, the weight of the correlation texture feature is saved as q2, the weight of the consistency texture feature is saved as q3, the weight of the angular second moment texture feature is saved as q4, the weight of the low frequency norm wavelet decomposition feature is saved as q5, the weight of the low frequency energy value wavelet decomposition feature is saved as q6, the weight of the horizontal high frequency norm wavelet decomposition feature is saved as q7, the weight of the horizontal high frequency energy value wavelet decomposition feature is saved as q8, the weight of the vertical high frequency norm wavelet decomposition feature is saved as q9, the weight of the vertical high frequency energy value wavelet decomposition feature is saved as q10, the weight of the diagonal high frequency norm wavelet decomposition feature is saved as q11, and the weight of the diagonal high frequency energy value wavelet decomposition feature is saved as q12.

[0041] Third step, the formula is used to establish the comprehensive texture feature index parameter of the specific crop, wherein represents the mean of the i-th texture feature parameter extraction result of the 20 samples.

[0042] Step 5: This step involves large-area identification and extraction of different types of crops in practical application, and the specific steps are as follows:

[0043] First step, using the standardized unmanned aerial vehicle flight scheme in step 1, the known N crops are standardized unmanned aerial vehicle flight experiments.

[0044] Second step, using step 2, the remote sensing image of each crop is standardized pretreatment operation, and the cropped crop gray scale image is obtained.

[0045] Third step, using step 3, the periodic structure scale of each crop standardized gray scale image is extracted, 20 random sample points are determined according to the scale of each crop, and the different types of texture feature parameters of sample points are calculated.

[0046] Fourth step, using step 4, the texture feature weight of each crop is calculated, and according to the weight and the texture feature extraction result of 20 random samples of each crop, the comprehensive texture feature index parameter of each crop is weighted calculated, and all known crop type comprehensive texture feature index parameter is saved as data set Y k ', wherein k is between 1 and N.

[0047] Fifth step, for unknown crops to be identified, using the processes of steps 1 to 3, different types of texture features of random samples of the crops are extracted, then according to the weight of different types of crops in the known database, the comprehensive texture feature index parameter Y" of the crops to be identified is calculated by formula , and the difference ratio Z between the crops to be identified and the known crops is calculated by formula , the calculation results of all known categories Z are sorted, and finally the crops to be identified are classified as the known crop type whose calculation result of Z is closest to 0.

[0048] In summary, the present application obtains the standardization of unmanned aerial vehicle remote sensing image at the same scale, and then determines the scale by the method of pixel statistical characteristics at this height. According to the scale, a rectangular window mask is established, the feature texture feature extraction is carried out in the mask area, including the extraction of gray level co-occurrence matrix texture feature and wavelet decomposition texture feature, and on this basis, the crop comprehensive texture feature index parameter representing the texture feature is weighted generated. Finally, the index parameter is used to realize the accurate identification of the type and distribution area of crops.

Claims

1. A method for accurate crop extraction based on drone remote sensing, characterized in that: The following steps are involved: Step 1: Acquisition of standard crop drone remote sensing images: Conduct standardized flight experiments on specific crop species using a multi-rotor drone, CCD high-definition camera, handheld GPS, compass, mobile computer, theodolite, and metal calibration plate. Determine the drone's flight altitude, flight parameters, and flight path, obtain calibrated drone remote sensing images at a fixed altitude, and perform standardized measurements on homogeneous remote sensing images of specific crops under these calibration conditions. Step 2: Standardized preprocessing of drone images: Use the coordinate data of the calibration points to perform geometric distortion correction on the drone-calibrated remote sensing images at a fixed altitude in Step 1. Perform geometric correction on the homogeneous remote sensing images of specific crops at that altitude. Convert the geometric correction results to grayscale, and crop the grayscale images of the specific crop remote sensing images to a uniform size to form a preprocessed result of a uniform standard size. Step 3: Standardized extraction of crop texture feature parameters: Generate random seed points for the preprocessed image of the specific crop in Step 2. Calculate statistical features within different sizes and determine the periodic scale of the crop with the seed points as the center. Generate 20 random mask regions based on the sizes, and extract gray-level co-occurrence matrix texture features from the specific crop image within the region, as well as wavelet texture features at different decomposition levels. Step 4. Establishment of comprehensive texture feature index parameters for crops: Calculate the texture feature weights of the gray-level co-occurrence matrix texture feature quantities and wavelet decomposition texture feature quantities in the 20 random mask areas in step 3, standardize each feature parameter based on the 20 samples, and then calculate the mean of a certain parameter after standardization. Then, based on the mean of different types, calculate the proportion of the mean to all values ​​as the weight, and on this basis establish the comprehensive texture feature index parameters for crops; Step 5. Identification and identification of crop species: Using the standardized UAV flight plan in step 1, perform standardized UAV coordinate positioning aerial photography flights on the entire area to be measured. Establish a window to traverse the standard pre-processed images of UAV remote sensing images and calculate different types of texture feature quantities. Establish identification rules based on the range of texture feature quantities and the comprehensive texture feature index parameters of crops to achieve accurate extraction of specific crops from the remote sensing images of the entire area and final determination of the spatial distribution area.

2. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: The step 1 is to determine the flight altitude, flight parameters and flight route of the drone, specifically by: selecting a determined square area on the map, measuring the area and average altitude of the area, and determining the latitude and longitude information of the boundary points of the area; dividing the area into multiple square overlapping sub-areas, and recording the center points of the sub-areas; calculating the optimal flight altitude based on the side length of the sub-area and the field of view angle of the drone CCD lens, and calculating the latitude and longitude coordinates of the center point of the sub-area by combining the latitude and longitude coordinates of the area corner points and the geometric position relationship of the sub-areas; placing a metal calibration plate above the center point of the sub-area, setting the actual flight altitude of the drone according to the altitude of the measurement area, setting the flight route of the drone and the hovering position for taking pictures, so that the drone flies along the straight path of the center point of the sub-area and sequentially measures and takes pictures.

3. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: In step 2, the grayscale image of the remote sensing image of the specific crop is cropped to a uniform size, and the grayscale remote sensing image is cropped using the position of the boundary points of the measurement area to delete the objects and images in the non-crop area.

4. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: The periodic scale of crops is determined in step 3, specifically by setting the number of random seed points and a variable neighborhood radius in the preprocessing result image for simulation, cyclically calculating the three first-order statistical characteristics of the mean, variance, and standard deviation in the neighborhood for each seed point, averaging the first-order statistical results in each neighborhood under all cycles, and obtaining a function of the statistical characteristics varying with the neighborhood radius. The double of the minimum radius value corresponding to the function graph in the stable state is the periodic structural scale value corresponding to the crops.

5. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: In step 3, the gray level co-occurrence matrix texture feature quantity of the specific crop image in the region is extracted, including extraction of contrast, correlation, consistency, and angular second-order moment.

6. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: In step 3, the coiflet-1 wavelet basis function is used to perform a 4-level wavelet decomposition process on the sub-image in the random training sample area, and the norm and energy value wavelet decomposition feature of low frequency, horizontal high frequency, vertical high frequency and diagonal high frequency are calculated according to the wavelet decomposition coefficients at different levels.

7. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: The texture feature weight is calculated in step 4, specifically by performing sum normalization processing on each texture feature parameter extraction result data set, calculating the mean of all sample points after the sum normalization processing of each texture feature quantity, calculating the sum of the normalized means of different types of texture features, and taking the ratio of the normalized mean of each texture feature to the sum as the weight of the texture feature quantity among all texture features.

8. The method for accurate crop extraction based on UAV remote sensing according to claim 1, characterized in that: In step 5, for unknown crops to be identified, different types of texture features of random samples of the crops are extracted, and the comprehensive texture feature index parameters of the crops to be identified are calculated based on the weights of different types of crops in the known database. The difference ratio between the crops to be identified and the known crops is calculated, and the crops to be identified are classified as the known crop type with the calculated difference ratio closest to 0.

Images