Crop yield prediction method and system based on unmanned aerial vehicle multi-spectral image fusion

Abstract

The application provides a crop yield prediction method and system based on unmanned aerial vehicle multi-spectral image fusion, and belongs to the yield prediction field.The crop yield prediction method comprises the following steps: collecting multi-spectral images of crops in a target area before the crops are mature and determining a crop yield prediction area; extracting gray images of five bands from the crop yield prediction area; determining the texture features of each band of each gray image; determining six texture indexes according to the texture features of each band; determining six vegetation indexes according to the average reflectivity of the five bands; determining six color indexes according to the average gray values of red, green and blue bands; determining a yield index according to the six texture indexes, the six vegetation indexes and the six color indexes; and determining the crop yield of the target area by using a random forest method according to the yield index. The image processing method reduces the influence on the crops, and the texture indexes, the vegetation indexes and the color indexes are fused, so that the yield prediction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of yield prediction, in particular to a crop yield prediction method and system based on unmanned aerial vehicle multi-spectral image fusion. BACKGROUND

[0002] Monitoring the planting area and yield of food crops has always been highly valued, and both the government and scientific researchers have been committed to studying how to timely and accurately understand and grasp information such as the planting yield of food crops. Timely acquisition of crop planting information can provide a scientific basis for the government to formulate agricultural production policies, which is of great significance to ensuring food security. In agricultural production, timely and accurate prediction of crop yield can also be of great significance for better implementation of crop management by farmers every year and the next year, especially in the determination of crop insurance, harvesting plans, storage needs, cash flow budgets, nutrition, pesticides, water, and other input decisions.

[0003] Traditional crop yield prediction methods need to measure row spacing, plant spacing, and also need to destructively sample to measure total kernel number per ear, seed setting rate, thousand kernel weight, and other parameters to finally calculate the predicted yield per mu. The method has high labor intensity, high cost, long time consumption, and strong subjectivity, mainly relying on the professional knowledge of the operator. The professional knowledge and experience of the operator vary, resulting in different measurement accuracy, which is difficult to maintain stability and reliability. The existing yield prediction methods using remote sensing technology mostly use only vegetation index or combine with ground crop physiological indicators. Such methods are only suitable for specific regions, varieties, and fertilization modes, but in actual application, the prediction accuracy may be insufficient due to the influence of year, region, variety, fertilization method, etc., which further leads to unreasonable arrangement of the number of harvesters and the allocation of personnel at the time of harvesting, resulting in low harvesting efficiency, resource waste, cost increase, and other problems. SUMMARY

[0004] The purpose of the present application is to provide a crop yield prediction method and system based on unmanned aerial vehicle multi-spectral image fusion, which can improve the accuracy of yield prediction and improve the harvesting efficiency.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] A crop yield prediction method based on unmanned aerial vehicle multi-spectral image fusion, comprising:

[0007] Collecting multi-spectral images of crops in the target area before maturity;

[0008] Determining a crop yield prediction area according to the multi-spectral images;

[0009] Extracting gray-scale images of 5 bands from the crop yield prediction area; the 5 bands include blue, green, red, red edge, and near-infrared bands;

[0010] According to each gray image, determine the texture feature of the corresponding wave band;

[0011] According to the texture feature of each wave band, determine 6 texture indexes;

[0012] Extract the reflectivity value of 5 wave bands from the crop yield estimation area, and determine the average reflectivity value of each wave band respectively;

[0013] According to the average reflectivity value of each wave band, determine 6 vegetation indexes;

[0014] Extract the gray value of red, green and blue wave bands from the crop yield estimation area, and determine the average gray value of red, green and blue wave bands respectively;

[0015] According to the average gray value of red, green and blue wave bands, determine 6 color indexes;

[0016] According to 6 texture indexes, 6 vegetation indexes and 6 color indexes, determine the yield index;

[0017] According to the yield index, use random forest method to determine the total yield of crops in the target area.

[0018] Optionally, the collection of the multispectral image of the target area before the maturity of crops specifically includes:

[0019] Collect multiple initial multispectral images of the target area by unmanned aerial vehicle carrying multispectral camera;

[0020] According to the characteristics of each initial multispectral image, stitch multiple initial multispectral images to obtain the multispectral image of the target area before the maturity of crops.

[0021] Optionally, the determination of the crop yield estimation area according to the multispectral image specifically includes:

[0022] Correct the multispectral image by the spectral correction plate placed in the target area in advance to obtain a multispectral correction image;

[0023] Remove the surrounding area in the multispectral correction image to obtain the crop yield estimation target area.

[0024] Optionally, the texture feature includes inverse gap texture, contrast texture, energy texture and autocorrelation texture;

[0025] Determine 6 texture indexes by using the following formula:

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032] wherein TI1, TI2, TI3, TI4, TI5, TI6 are six texture indexes, CON Red is the contrast texture of the red band, ENENIR is the energy texture of the near-infrared band, ENE Red is the energy texture of the red band, COR Red is the autocorrelation texture of the red band, CON NIR is the contrast texture of the near-infrared band.

[0033] Optionally, the following formula is used to determine the six vegetation indexes:

[0034] VI1 = (RE-NIR) / (B-R) ;

[0035] VI2 = R-B ;

[0036] VI3 = (NIR-R) / (NIR+R) ;

[0037] VI4 = NIR / B ;

[0038] VI5 = 1.5*(R-B) / (R+B+0.5) ;

[0039] VI6 = 1.16*(NIR-R) / (NIR+R+0.16) ;

[0040] wherein VI1, VI2, VI3, VI4, VI5, VI6 are six vegetation indexes, RE is the average reflectance of the red edge band, NIR is the average reflectance of the near-infrared band, B is the average reflectance of the blue band, and R is the average reflectance of the red band.

[0041] Optionally, the six color indexes are determined according to the average gray values of the red, green, and blue bands, and specifically include:

[0042] The average gray value of each band is normalized to obtain the normalized gray value of the corresponding band;

[0043] The six color indexes are determined according to the normalized gray values of the bands.

[0044] Optionally, the following formula is used to determine the normalized gray values of the red, green, and blue bands, respectively:

[0045] r=R / (R+G+B);

[0046] g=G / (R+G+B);

[0047] b=B / (R+G+B);

[0048] Wherein, r is the normalized gray value of the red wave band, g is the normalized gray value of the green wave band, b is the normalized gray value of the blue wave band, R is the average gray value of the red wave band, G is the average gray value of the green wave band, and B is the average gray value of the blue wave band.

[0049] Optionally, the following formula is used to determine the six color indexes:

[0050] CI1=2g-b-r;

[0051] CI2=(g 2 -r 2 ) / (g 2 +r 2 );

[0052] CI3=(g 2 -b*r) / (g 2 +b*r);

[0053] CI4=(r-g) / (r+g-b);

[0054] CI5=3g-2.4r-b;

[0055] CI6=(2g-b-r) / (2g+b+r);

[0056] Wherein CI1, CI2, CI3, CI4, CI5, and CI6 are the newly established six color indexes, g is the normalized gray value of the green wave band, b is the normalized gray value of the blue wave band, and r is the normalized gray value of the red wave band.

[0057] Optionally, the total yield of crops in the target area is determined according to the yield index and by using a random forest method, and specifically includes:

[0058] The yield per mu of crops in the target area is determined according to the yield index and based on a machine learning random forest model.

[0059] The total yield of crops is determined according to the crop planting area and the yield per mu of crops in the target area.

[0060] To achieve the above-mentioned purposes, the present application also provides the following solutions:

[0061] A crop yield prediction system based on unmanned aerial vehicle multi-spectral image fusion, comprising:

[0062] An image acquisition module is configured to acquire a multi-spectral image of a target area before the crops in the target area mature.

[0063] A crop yield area determination module is connected to the image acquisition module and configured to determine a crop yield area according to the multi-spectral image.

[0064] A gray image extraction module is connected to the crop yield area determination module and configured to extract a gray image of five wave bands from the crop yield area, wherein the five wave bands include blue, green, red, red edge and near-infrared wave bands.

[0065] A texture feature determination module is connected to the gray image extraction module and configured to determine a texture feature of each wave band according to each gray image.

[0066] A texture index determination module is connected to the texture feature determination module and configured to determine six texture indexes according to the texture features of the wave bands.

[0067] A reflectivity determination module is connected to the crop yield area determination module and configured to extract reflectivity values of the five wave bands from the crop yield area and determine an average reflectivity value of each wave band.

[0068] A vegetation index determination module is connected to the reflectivity determination module and configured to determine six vegetation indexes according to the average reflectivity values of the wave bands.

[0069] A gray value determination module is connected to the crop yield area determination module and configured to extract gray values of red, green and blue wave bands from the crop yield area and determine average gray values of the red, green and blue wave bands.

[0070] A color index determination module is connected to the gray value determination module and configured to determine six color indexes according to the average gray values of the red, green and blue wave bands.

[0071] A yield index determination module is connected to the texture index determination module, the vegetation index determination module and the color index determination module and configured to determine a yield index according to the six texture indexes, the six vegetation indexes and the six color indexes.

[0072] A crop yield determination module is connected to the yield index determination module and configured to determine a total crop yield of the target area according to the yield index by using a random forest method.

[0073] According to the specific embodiments of the present application, the following technical effects are disclosed: a multispectral image before the maturity of crops is collected, texture indexes, vegetation indexes and color indexes of crop canopies in the multispectral image are extracted respectively, the texture indexes, the vegetation indexes and the color indexes are used to determine the crop yield of a target area by using a random forest method, the crop yield is predicted by using an image processing method, the influence on crops is reduced, the texture indexes, the vegetation indexes and the color indexes are fused, the characteristics of the crop canopy are fully extracted, and the prediction accuracy of the yield is improved. BRIEF DESCRIPTION OF DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0075] Figure 1 The flowchart of the crop yield prediction method based on the unmanned aerial vehicle multispectral image fusion of the present application;

[0076] Figure 2 The position diagram of the spectral correction plate;

[0077] Figure 3 The module structure diagram of the crop yield prediction system based on the unmanned aerial vehicle multispectral image fusion of the present application.

[0078] Symbol explanation:

[0079] Image acquisition module-1, crop yield determination area determination module-2, gray image extraction module-3, texture feature determination module-4, texture index determination module-5, reflectivity determination module-6, vegetation index determination module-7, gray value determination module-8, color index determination module-9, yield index determination module-10, crop yield determination module-11, spectral correction plate-a, b, c, d. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0081] The application aims to provide a crop yield prediction method and system based on unmanned aerial vehicle multi-spectral image fusion, which reduces the influence on crops by predicting crop yield in an image processing manner, and fully extracts the characteristics of the crop canopy by fusing 6 texture indexes, 6 vegetation indexes and 6 color indexes, thereby improving the yield prediction accuracy.

[0082] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below in combination with the drawings and specific embodiments.

[0083] As shown in Figure 1 The application provides a crop yield prediction method based on unmanned aerial vehicle multi-spectral image fusion, which comprises the following steps:

[0084] S1: collecting multi-spectral images of crops before maturation in a target area.

[0085] Specifically, a plurality of initial multi-spectral images of the target area are collected by a multi-spectral camera carried by an unmanned aerial vehicle. According to the characteristics of each initial multi-spectral image, the plurality of initial multi-spectral images are spliced to obtain the multi-spectral images of crops before maturation in the target area.

[0086] The application uses an unmanned aerial vehicle platform to collect spectral images of the crop canopy, which is fast and time-saving, and can estimate the yield of a large area of crops without personnel entering the paddy field, thereby reducing labor intensity and operation cost and improving operation efficiency.

[0087] S2: determining a crop yield estimation area according to the multi-spectral images.

[0088] Specifically, the multi-spectral images are corrected by a spectral correction plate placed in the target area in advance to obtain multi-spectral correction images. The peripheral area in the multi-spectral correction images is removed to obtain a crop yield estimation target area. In this embodiment, a plurality of spectral correction plates are placed in the target area before the unmanned aerial vehicle collects information by aerial photography.

[0089] S3: extracting 5-band gray images from the crop yield estimation area; the 5 bands include blue, green, red, red edge and near-infrared bands. Specifically, according to the correlation between the crop canopy images of each band and the crop growth status and yield, the gray images of the blue, green, red, red edge and near-infrared bands in the 435nm-870nm band range are screened out.

[0090] S4: determining the texture features of the corresponding bands according to each gray image.

[0091] S5: determining 6 texture indexes according to the texture features of each band.

[0092] Specifically, the texture features include inverse distance texture, contrast texture, energy texture and autocorrelation texture. In the embodiment, the texture indices corresponding to any two different texture features are calculated by using the following formula ; wherein, TI is the texture index corresponding to T1 and T2, and T1 and T2 are any two different texture features. Then, according to the correlation with the yield data, the first six texture indices TI1, TI2, TI3, TI4, TI5 and TI6 with high correlation are screened out:

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099] ; wherein, TI1, TI2, TI3, TI4, TI5 and TI6 are the six texture indices, CON Red is the contrast texture of the red band, ENE NIR is the energy texture of the near-infrared band, ENE Red is the energy texture of the red band, COR Red is the autocorrelation texture of the red band, CON NIR is the contrast texture of the near-infrared band.

[0100] S6: Extract the reflectance values of the five bands from the crop yield estimation area, and determine the reflectance average value of each band respectively. Specifically, the reflectance values of different bands of each pixel point in the crop yield estimation area are extracted. According to the correlation between the spectral values of each band of the crop canopy and the crop growth status and yield, the reflectance average values of the blue, green, red, red edge and near-infrared five bands in the 435nm-870nm band range are screened out.

[0101] S7: Determine six vegetation indices according to the reflectance average values of each band. Specifically, the following formula is used to determine the six vegetation indices:

[0102] VI1=(RE-NIR) / (B-R);

[0103] VI2=R-B;

[0104] VI3=(NIR-R) / (NIR+R);

[0105] VI4=NIR / B;

[0106] VI5=1.5*(R-B) / (R+B+0.5);

[0107] VI6=1.16*(NIR-R) / (NIR+R+0.16);

[0108] wherein VI1, VI2, VI3, VI4, VI5, VI6 are six vegetation indices, RE is the average reflectance of the red edge band, NIR is the average reflectance of the near-infrared band, B is the average reflectance of the blue band, and R is the average reflectance of the red band.

[0109] S8: Extract the gray values of the red, green, and blue bands from the crop yield estimation area, and determine the average gray DN values of the red, green, and blue bands, respectively. Specifically, the aerial image contains an RGB image and multiple single-band gray images. This step extracts the gray values of the red, green, and blue bands from the crop yield estimation area in the RGB image.

[0110] S9: Determine six color indices according to the average gray values of the red, green, and blue bands.

[0111] Further, step S9 specifically includes:

[0112] S901: Normalize the average gray values of each band to obtain the normalized gray values of the corresponding band. Specifically, the following formula is used to determine the normalized gray values of the red, green, and blue bands, respectively:

[0113] r=R / (R+G+B);

[0114] g=G / (R+G+B);

[0115] b=B / (R+G+B);

[0116] wherein r is the normalized gray value of the red band, g is the normalized gray value of the green band, and b is the normalized gray value of the blue band.

[0117] S902: Determine six color indices according to the normalized gray values of each band. Specifically, first calculate the spectral values of any band among the red, green, and blue bands, then calculate the color indices by combining any three bands, and finally select the top six color indices with high correlation with yield data according to the correlation:

[0118] CI1=2g-b-r;

[0119] CI2=(g 2 -r 2 ) / (g2 + 2 ;

[0120] CI3 = (g 2 -b*r) / (g 2 +b*r);

[0121] CI4 = (r-g) / (r+g-b);

[0122] CI5 = 3g-2.4r-b;

[0123] CI6 = (2g-b-r) / (2g+b+r);

[0124] wherein CI1, CI2, CI3, CI4, CI5, CI6 are six new color indices, g is the normalized gray value of the green wave band, b is the normalized gray value of the blue wave band, and r is the normalized gray value of the red wave band.

[0125] S10: determining a yield index according to the six texture indices, the six vegetation indices, and the six color indices.

[0126] S11: determining the total yield of crops in the target area according to the yield index by using a random forest method.

[0127] Further, step S11 specifically comprises:

[0128] determining the yield per mu of crops in the target area based on a machine learning random forest model according to the yield index. Specifically, the yield per mu of crops is determined by using the following formula:

[0129] Yield = fRF(TI1, TI2, TI3, TI4, TI5, TI6, VI1, VI2, VI3, VI4, VI5, VI6, CI1, CI2, CI3, CI4, CI5, CI6); wherein Yield is the yield per mu of crops, f RF () is a random forest RF model.

[0130] determining the total yield of crops according to the crop planting area and the yield per mu of crops in the target area. Specifically, the total yield of crops is determined by using the formula W = Yield*N; wherein W is the total yield of crops, Yield is the yield per mu of crops, and N is the crop planting area in the target area.

[0131] In steps S7 and S9, the first six vegetation indices and the first six color indices with high correlation with the yield data are screened out according to the correlation.

[0132] The unmanned aerial vehicle flies at a height of 20-150 m when the image is captured in step S1. The forward repeat rate of the unmanned aerial vehicle is greater than or equal to 60%, the lateral repeat rate of the unmanned aerial vehicle is greater than or equal to 55%, and the flight speed is less than or equal to 10 m / s. During the flight of the unmanned aerial vehicle, the lens direction of the multispectral camera is perpendicular to the ground. The trigger signal is transmitted to the multispectral camera at equal intervals by the flight control system of the unmanned aerial vehicle, or the multispectral camera takes pictures at fixed time intervals. In addition, the position and attitude recorder is triggered at the same time to collect geographic information such as GPS. According to the same features of the repeated images in the front and back and left and right directions, the images are spliced to obtain the multispectral image of the target area before the crops mature.

[0133] In step S2, the multispectral image is corrected by the spectral correction board placed in the target area in advance to obtain a multispectral correction image, which specifically includes: correcting the multispectral image according to the calibration correction coefficient of the spectral correction board to obtain the reflectivity of each pixel point in the multispectral image. Specifically, the calibration correction coefficient of the spectral correction board refers to the reflectivity of each spectral correction board, and the reflectivity of each spectral correction board is known and can be obtained in advance: the reflected light intensity E2 of each spectral correction board is measured under the standard light of known light intensity E1, and the reflectivity corresponding to each spectral correction board is Ri=E1 / E2, that is, the calibration board correction coefficient.

[0134] For each spectral correction board, the average spectral value of all pixel points in the spectral correction board is taken as the abscissa, and the calibration correction coefficient is taken as the ordinate to form the coordinate point corresponding to the spectral correction board. For example, four coordinate points corresponding to four spectral correction boards can be obtained. Linear fitting is performed according to the coordinate points corresponding to all spectral correction boards to obtain a reflectivity correction function between the spectrum and the reflectivity. Further, the reflectivity of each pixel point in the multispectral image is obtained by using the reflectivity correction function according to the spectral value of each pixel point in the multispectral image, and the spectral correction is completed.

[0135] Since the spectral values of the crop yield estimation target area and the surrounding area are different, the multispectral correction image is segmented to remove the redundant parts such as roads and ridges to obtain an accurate crop yield estimation target area.

[0136] In order to better understand the scheme of the present application, the following will be further described in conjunction with specific embodiments.

[0137] This embodiment takes rice yield estimation as an example to introduce the process of the crop yield prediction method based on unmanned aerial vehicle multispectral image fusion of the present application.

[0138] First step: Place spectral correction boards with known reflectivity in the rice yield estimation target area. For example, Figure 2As shown, in the case of clear and few cloudy weather illumination, first place 4 known reflectivity spectral correction plates a, b, c, d at the target rice field edge.

[0139] Second step: One week before the mature harvest of rice, use the unmanned aerial vehicle carrying the multi-spectral imaging system to take pictures, and collect multiple initial multi-spectral images of rice. During the flight of the unmanned aerial vehicle, the direction of the multi-spectral camera lens is perpendicular to the ground. When collecting image information, the heading repetition rate of the unmanned aerial vehicle is set to not less than 60%, the lateral repetition rate of the unmanned aerial vehicle is set to not less than 55%, and the flight speed is not greater than 10 m / s. The trigger signal is sent to the camera by the flight control system of the unmanned aerial vehicle at equal intervals, or the trigger signal is sent to the camera at fixed time intervals, and the multi-spectral camera and the position and attitude recorder are triggered to take pictures and collect GPS geographic information at the same time.

[0140] Third step: According to the image features, the initial multi-spectral images are spliced to obtain the multi-spectral image of the rice. According to the same features of the repeated images before and after the heading and the repeated images on the left and right sides, the image splicing is completed.

[0141] Fourth step: The multi-spectral image is respectively processed by image distortion correction and spectral correction through the spectral correction plate.

[0142] Fifth step: The multi-spectral image is segmented to segment out the rice yield target area. Specifically, the multi-spectral image is segmented to remove the redundant parts such as roads and ridges by using the different spectral values of the rice yield target area and the surrounding area.

[0143] Sixth step: Extract the blue, green, red, red edge, and near-infrared five-band gray-scale images of the rice canopy in the rice yield target area.

[0144] Seventh step: Calculate the texture features of four gray-level co-occurrence matrices of inverse distance, contrast, energy, and autocorrelation for the blue, green, red, red edge, and near-infrared five-band gray-scale images.

[0145] Eighth step: Six texture indices are newly created according to the four texture features. Specifically, six texture indices are newly created according to the four texture features:

[0146]

[0147]

[0148]

[0149]

[0150]

[0151]

[0152] wherein P is the Pearson correlation coefficient.

[0153] Step 9: Extract the reflectance values of the blue, green, red, red edge, and near-infrared bands and the reflectance average value in the rice yield estimation target area.

[0154] Step 10: Use the reflectance average values of the 5 characteristic bands to create 6 new vegetation indices.

[0155] Specifically, the reflectance average values of the blue, green, red, red edge, and near-infrared bands in the 435nm-870nm band range are extracted. The 5 characteristic bands selected above are used to create 6 new vegetation indices:

[0156] VI1 = (RE-NIR) / (B-R) (P = 0.7742);

[0157] VI2 = R-B (P = 0.5921);

[0158] VI3 = (NIR-R) / (NIR+R) (P = 0.6741);

[0159] VI4 = NIR / B (P = 0.6432);

[0160] VI5 = 1.5*(R-B) / (R+B+0.5) (P = 0.8423);

[0161] VI6 = 1.16*(NIR-R) / (NIR+R+0.16) (P = 0.6402).

[0162] Step 11: Extract the gray values of the red, green, and blue bands in the rice yield estimation target area, and determine the average DN values of the red, green, and blue bands respectively.

[0163] Step 12: Use the average DN values of the red, green, and blue bands to create 6 new color indices.

[0164] Step 13: Merge the 6 texture indices, 6 vegetation indices, and 6 color indices to construct yield indices. Use the machine learning random forest model to predict the yield per mu of rice, and calculate the total yield of rice according to the target area rice planting area and the predicted yield per mu of rice.

[0165] The final results show that the model determination coefficient (P 2) is 0.84, the mean absolute error (MAE) is 47.7 kg / acre, and the mean absolute percentage error (MAPE) is 7.86%. As can be seen, the present application fully taps the optical data of the crop canopy, fuses the yield index by using the reconstructed vegetation index, color index and self-defined texture index, and is applicable to crops of different regions, different varieties and different fertilization modes, and the measurement process is fixed and streamlined, human influence is excluded, and the universality and accuracy of crop yield prediction are improved.

[0166] As shown in Figure 3 The crop yield prediction system based on unmanned aerial vehicle multi-spectral image fusion of the present application comprises an image acquisition module 1, a crop yield determination area determination module 2, a gray image extraction module 3, a texture feature determination module 4, a texture index determination module 5, a reflectivity determination module 6, a vegetation index determination module 7, a gray value determination module 8, a color index determination module 9, a yield index determination module 10 and a crop yield determination module 11.

[0167] The image acquisition module 1 is used to acquire multi-spectral images of crops before maturation in the target area.

[0168] Specifically, the image acquisition module 1 comprises an initial image acquisition unit and a splicing unit. The initial image acquisition unit is used to acquire a plurality of initial multi-spectral images of the target area by a multi-spectral camera carried by an unmanned aerial vehicle. The splicing unit is connected with the initial image acquisition unit, and the splicing unit is used to splice the plurality of initial multi-spectral images according to the characteristics of each initial multi-spectral image to obtain the multi-spectral image of the crops before maturation in the target area.

[0169] The crop yield determination area determination module 2 is connected with the image acquisition module 1, and the crop yield determination area determination module 2 is used to determine the crop yield determination area according to the multi-spectral image.

[0170] Specifically, the crop yield determination area determination module 2 comprises a correction unit and a region extraction unit. The correction unit is connected with the image acquisition module 1, and the correction unit is used to correct the multi-spectral image by a spectral correction plate placed in the target area in advance to obtain a multi-spectral correction image. The region extraction unit is connected with the correction unit, and the region extraction unit is used to remove the surrounding region in the multi-spectral correction image to obtain the crop yield determination target area.

[0171] The gray image extraction module 3 is connected with the crop yield determination area determination module 2, and the gray image extraction module 3 is used to extract a gray image of 5 wave bands from the crop yield determination area; the 5 wave bands include blue, green, red, red edge and near-infrared wave bands.

[0172] The texture feature determination module 4 is connected with the gray image extraction module 3, and the texture feature determination module 4 is configured to determine the texture feature of each wave band according to each gray image.

[0173] The texture index determination module 5 is connected with the texture feature determination module 4, and the texture index determination module 5 is configured to determine 6 texture indexes according to the texture features of each wave band.

[0174] The reflectivity determination module 6 is connected with the crop yield determination area determination module 2, and the reflectivity determination module 6 is configured to extract the reflectivity values of 5 wave bands from the crop yield determination area and determine the average reflectivity of each wave band.

[0175] The vegetation index determination module 7 is connected with the reflectivity determination module 6, and the vegetation index determination module 7 is configured to determine 6 vegetation indexes according to the average reflectivity of each wave band.

[0176] The gray value determination module 8 is connected with the crop yield determination area determination module 2, and the gray value determination module 8 is configured to extract the gray values of red, green and blue wave bands from the crop yield determination area and determine the average gray values of the red, green and blue wave bands.

[0177] The color index determination module 9 is connected with the gray value determination module 8, and the color index determination module 9 is configured to determine 6 color indexes according to the average gray values of the red, green and blue wave bands.

[0178] Specifically, the color index determination module 9 includes a normalization unit and a color determination unit. The normalization unit is connected with the gray value determination module 8, and the normalization unit is configured to normalize the average gray value of each wave band to obtain the normalized gray value of the corresponding wave band. The color index determination unit is connected with the normalization unit, and the color index determination unit is configured to determine 6 color indexes according to the normalized gray value of each wave band.

[0179] The yield index determination module 10 is connected with the texture index determination module 5, the vegetation index determination module 7 and the color index determination module 9 respectively, and the yield index determination module 10 is configured to determine the yield index according to the 6 texture indexes, the 6 vegetation indexes and the 6 color indexes.

[0180] The crop yield determination module 11 is connected with the yield index determination module 10, and the crop yield determination module 11 is configured to determine the total yield of the target area by using the random forest method according to the yield factor.

[0181] Specifically, the crop yield determination module 11 includes a crop yield per acre determination unit and a total yield determination unit. The crop yield per acre determination unit is connected to the yield index determination module 10, and is used to determine the crop yield per acre in the target area based on the yield index and a machine learning random forest model. The total yield determination unit is connected to the crop yield per acre determination unit, and is used to determine the total crop yield based on the crop planting area and crop yield per acre in the target area.

[0182] Compared to existing technologies, the crop yield prediction system based on UAV multispectral image fusion of the present invention has the same beneficial effects as the crop yield prediction method based on UAV multispectral image fusion described above, and will not be repeated here.

[0183] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0184] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A crop yield prediction method based on unmanned aerial vehicle multi-spectral image fusion, characterized in that, The crop yield prediction method based on unmanned aerial vehicle multi-spectral image fusion comprises the following steps: Collecting multi-spectral images of crops in a target area before the crops are mature; Determining a crop yield survey area according to the multi-spectral images; Extracting gray-scale images of five wave bands from the crop yield survey area, wherein the five wave bands include blue, green, red, red edge and near-infrared wave bands; Determining texture features of corresponding wave bands according to each gray-scale image, wherein the texture features include inverse distance texture, contrast texture, energy texture and autocorrelation texture; Determining six texture indexes according to the texture features of the wave bands by using the following formula: Wherein, TI1, TI2, TI3, TI4, TI5, TI6 are 6 texture indexes, CON Red is the contrast texture of the red band, ENE NIR is the energy texture of the near-infrared band, ENE Red is the energy texture of the red band, CON Red is the autocorrelation texture of the red band, CON NIR is the contrast texture of the near-infrared band; Extracting reflectivity values of the five wave bands from the crop yield survey area, and determining average reflectivity values of each wave band respectively; Determining six vegetation indexes according to the average reflectivity values of the wave bands: VI1=(RE-NIR) / (B-R); VI2=R-B; VI3=(NIR-R) / (NIR+R); VI4=NIR / B; VI5=1.5*(R-B) / (R+B+0.5); VI6=1.16*(NIR-R) / (NIR+R+0.16); wherein RE is the average reflectivity value of the red edge wave band, NIR is the average reflectivity value of the near-infrared wave band, B is the average reflectivity value of the blue wave band, and R is the average reflectivity value of the red wave band; Extracting gray-scale values of the red, green and blue wave bands from the crop yield survey area, and determining average gray-scale values of the red, green and blue wave bands respectively; Normalizing the average gray-scale values of each wave band to obtain normalized gray-scale values of the corresponding wave bands: r=R / (R+G+B); g=G / (R+G+B); b=B / (R+G+B); wherein r is the normalized gray-scale value of the red wave band, g is the normalized gray-scale value of the green wave band, b is the normalized gray-scale value of the blue wave band, R is the average gray-scale value of the red wave band, G is the average gray-scale value of the green wave band, and B is the average gray-scale value of the blue wave band; Determining six color indexes according to the normalized gray-scale values of the wave bands by using the following formula: CI1=2g-b-r; CI2= (g 2 -r 2 ) / (g 2 +r 2 ); CI3 = (g 2 -b*r) / (g 2 +b*r); CI4=(r-g) / (r+g-b); CI5=3g-2.4r-b; CI6=(2g-b-r) / (2g+b+r); wherein g is the normalized gray-scale value of the green wave band, b is the normalized gray-scale value of the blue wave band, and r is the normalized gray-scale value of the red wave band; Determining a yield index according to the six texture indexes, the six vegetation indexes and the six color indexes; According to the yield index, determine the crop yield per mu of the target area based on a machine learning random forest model: Yield=f RF (TI1,TI2,TI3,TI4,TI5,TI6,VI1,VI2,VI3,VI4,VI5,VI6,CI1,CI2,CI3,CI4,CI5,CI6); wherein Yield is the crop yield per mu, f RF () is a random forest RF model, VI1, VI2, VI3, VI4, VI5, VI6 are 6 vegetation indices, CI1, CI2, CI3, CI4, CI5, CI6 are 6 newly built color indices; Determining a total crop yield W=Yield*N according to the crop planting area of the target area and the crop yield per mu; wherein W is the total crop yield, Yield is the crop yield per mu, and N is the crop planting area of the target area. 2.The crop yield prediction method based on unmanned aerial vehicle multi-spectral image fusion according to claim 1, characterized in that, The method for collecting multi-spectral images of crops in a target area before the crops are mature comprises the following steps: Collecting multiple initial multi-spectral images of the target area by an unmanned aerial vehicle carrying a multi-spectral camera; Splicing the multiple initial multi-spectral images according to the characteristics of the initial multi-spectral images to obtain the multi-spectral images of the crops in the target area before the crops are mature. 3.The crop yield prediction method based on unmanned aerial vehicle multi-spectral image fusion according to claim 1, characterized in that, The method for determining a crop yield survey area according to the multi-spectral images comprises the following steps: The multispectral image is corrected by a spectral correction plate placed in the target area in advance to obtain a multispectral correction image; The peripheral area in the multispectral correction image is removed to obtain a crop yield estimation target area.

4. A crop yield prediction system based on unmanned aerial vehicle multispectral image fusion, applied to the crop yield prediction method based on unmanned aerial vehicle multispectral image fusion in any one of claims 1-3, characterized in that, The crop yield prediction system based on unmanned aerial vehicle multispectral image fusion comprises: An image acquisition module is configured to acquire a multispectral image of a target area before the crops are matured; A crop yield estimation area determination module is connected to the image acquisition module and configured to determine a crop yield estimation area according to the multispectral image; A gray image extraction module is connected to the crop yield estimation area determination module and configured to extract a gray image of five wave bands from the crop yield estimation area; the five wave bands include blue, green, red, red edge and near-infrared wave bands; A texture feature determination module is connected to the gray image extraction module and configured to determine a texture feature of a corresponding wave band according to each gray image; A texture index determination module is connected to the texture feature determination module and configured to determine six texture indexes according to the texture features of the wave bands; A reflectivity determination module is connected to the crop yield estimation area determination module and configured to extract reflectivity values of the five wave bands from the crop yield estimation area and determine an average reflectivity value of each wave band; A vegetation index determination module is connected to the reflectivity determination module and configured to determine six vegetation indexes according to the average reflectivity values of the wave bands; A gray value determination module is connected to the crop yield estimation area determination module and configured to extract gray values of red, green and blue wave bands from the crop yield estimation area and determine average gray values of the red, green and blue wave bands; A color index determination module is connected to the gray value determination module and configured to determine six color indexes according to the average gray values of the red, green and blue wave bands; A yield index determination module is connected to the texture index determination module, the vegetation index determination module and the color index determination module and configured to determine a yield index according to the six texture indexes, the six vegetation indexes and the six color indexes; A crop yield determination module is connected to the yield index determination module and configured to determine a total crop yield of the target area according to the yield index by using a random forest method.

Images